Received June 7, 2020, accepted June 23, 2020, date of publication July 3, 2020, date of current version July

16, 2020.
Digital Object Identifier 10.1109/ACCESS.2020.3006887

A Comprehensive Study on IoT Based Accident

Detection Systems for Smart Vehicles
UNAIZA ALVI1 , MUAZZAM A. KHAN KHATTAK 2 , (Senior Member, IEEE),
BALAWAL SHABIR 1 , ASAD WAQAR MALIK 1 , (Senior Member, IEEE),
AND SHER RAMZAN MUHAMMAD 3
1 Department of Computing, School of Electrical Engineering and Computer Science (SEECS), National University of Sciences and Technology (NUST),
Islamabad 44000, Pakistan
2 Department of Computer Science, Quaid-i-Azam University, Islamabad 45320, Pakistan
3 Faculty of Computing and Information Technology (FCIT), King Abdulaziz University, Jeddah 21589, Saudi Arabia

Corresponding author: Muazzam A. Khan Khattak ([email protected])

ABSTRACT With population growth, the demand for vehicles has increased tremendously, which has
created an alarming situation in terms of traffic hazards and road accidents. The road accidents percentage
is growing exponentially and so are the fatalities caused due to accidents. However, the primary cause of
the increased rate of fatalities is due to the delay in emergency services. Many lives could be saved with
efficient rescue services. The delay happens due to traffic congestion or unstable communication to the
medical units. The implementation of automatic road accident detection systems to provide timely aid is
crucial. Many solutions have been proposed in the literature for automatic accident detection. The techniques
include crash prediction using smartphones, vehicular ad-hoc networks, GPS/GSM based systems, and
various machine learning techniques. With such high rates of deaths associated with road accidents, road
safety is the most critical sector that demands significant exploration. In this paper, we present a critical
analysis of various existing methodologies used for predicting and preventing road accidents, highlighting
their strengths, limitations, and challenges that need to be addressed to ensure road safety and save valuable
lives.

INDEX TERMS GSM, GPS, accident detection, IoT, smart cities.

NOMENCLATURE DDRL Drowsiness Detection using Representation

AC Alcohol Check Learning
Acc Accuracy DIDCNN Distraction Identification with Deep Convo-
Accel Accelerometer lutional Neural Network
ADANN Accident Detection using Artificial Neural DLC Driver License Check
Network FPR False Positive Rate
ADFL Accident Detection using Fuzzy Logic H High
AL Alcohol IR Infrared Sensor
APSVM Accident Prediction using Support Vector L Low
Machine LS Limit Switches
AR Automatic Reporting M Medium
D Distance MC Micro-controller
DBA Driver Behaviour Analysis ML/AI Machine Learning/Artificial Intelligence
DBT Deceleration Based on Traffic signals N No
DC Drowsiness Check N/A Not Applicable
DDDD Drowsy and Drunk Driving Detection P Pressure
PDCNN Pedestrian Detection with Convolutional
Neural Network
The associate editor coordinating the review of this manuscript and RF Radio Frequency
approving it for publication was Alessandra De Benedictis. RP Rasberry Pi

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
122480 VOLUME 8, 2020
U. Alvi et al.: Comprehensive Study on IoT Based Accident Detection Systems for Smart Vehicles

S Speed the response time after an accident has occurred. Signifi-

SB Seat Belt cant research has been carried out to address this issue and
SBC Seat Belt Check to minimize the response time following an accident [1].
SC Speed Check Different approaches are used for this purpose. In this context,
SDO Safe distance from Obstacles VANET (Vehicular Ad-hoc Network) can be utilized [8], [9],
SH Smart Helmet in which every moving vehicle acts as a node. On occurrence
STF Switch to Terminate False alarm of accident, the alert messages are communicated via RF
Temp Temperature (Radio Frequency) module [10]. One approach uses limit
UAS Using Accelerometer Sensors switches to detect an accident, GSM (Global System for
UCSA Using Cell phone, Speed and Accelerometer Mobile Communications) is used to send an alert message
sensors and location of accident is traced by GPS (Global Positioning
UDS Using Distance Sensors System) module [11]. Smartphone based systems that use
UGG Using GPS and GSM an android app to detect vehicle crash are also proposed.
UGGARFT Using Gyroscope, GPS, android app and RF These systems measure change of tilt angle by means of an
transmitter accelerometer sensor, speed by means of GPS and send an
ULS Using Limit Switches alert on detection of accident [12]. Some systems focus on
UPS Using Piezoelectric Sensors preventive strategy because at the end, goal is to save lives.
UPST Using Pressure Sensor and Tilt angle This system particularly focuses on the safety of two wheelers
USSG Using Shock Sensors and GPS and checks if the driver is drowsy [2]. One technique focuses
UVS Using Vibration Sensors on using the accelerometer sensor, by monitoring the vehicle
V Vibration speed and report an accident as it reaches below the threshold
VBA Vehicle Behaviour Analysis point [13]. Another approach aims on preventing drunken
Y Yes driving by installing alcohol sensors on the steering wheel
and not allowing the driver to drive if he’s intoxicated by
I. INTRODUCTION measuring the alcohol content in his oxygen [3], [14].
Due to rapid growth of world population, the demand for In this approach, data is obtained from various sensors and
vehicles has increased tremendously, resultantly problems of event logs to extract prominent features for collision detection
traffic congestion and road accidents has also increased. The model. Various intelligent computational models are used
general population’s life is under high risk, if any accident to detect accidents. These models include nearest neighbor,
occurs there’s a long reaction time which increments the neural networks and regression trees [15]. Vehicle behaviors
number of deaths, therefore an automatic accident detection can be analyzed given its position and velocity values, and can
system must exist to overcome this situation. Statistics show be helpful in the detection of accidents. To detect an accident
that leading cause of death by injury is road accidents [1]. and to distinguish it from normal cases, different machine
There can be multiple causes of road accidents, some of learning algorithms like Support Vector Machine (SVM),
them are, driver negligence due to drowsiness [2], driving Artificial Neural Network (ANN) and Random Forests (RF)
while intoxicated [3], over speeding [4], [5] etc. Some studies are implemented on traffic data [16]. Machine learning tech-
show that weather conditions can also contribute towards the niques can also be utilized to determine the severity of
severity of an accident such as fog, rain, high winds. High accident. Different algorithms such as k-means clustering,
winds can directly influence the vehicle which may deviate SVM under reinforcement learning by fetching real time
the vehicle from road, or indirectly due to obstruction dangers data like velocity which is obtained by means of vibration
present on the roads such as trees, walls etc., [6]. sensors installed in vehicle and distance which is obtained
Road crashes can be seen as a collision between any on- by means of ultrasonic sensors [17]. Accidents can also be
road vehicles, obstacles or pedestrians. The survival rate of detected by pre-trained surveillance cameras installed on the
victim is highly reliant on how long an ambulance takes to highways [18], as illustrated in Figure 1. Another machine
reach the site of the accident and then carry the patient to learning approach uses fuzzy logic. Data like number of
the hospital. In most cases of road accidents, the injuries are vehicles in each zone, speed of cars in particular lanes etc.
not severe and the life of the victim can be rescued, however is collected and then decisions are made accordingly. In this
due to late arrival of the rescue teams, the injuries turn deadly. technique, a situation is detected as an accident whenever
Thus, the main goal is to identify where the accident occurred, there’s some sort of disturbance in surrounding lanes [19].
send the information to the rescue teams in considerably less Many systems focused on taking preventive measures,
time, so that they can take the necessary actions, to save the along with detection to avoid accidents such as controlling the
life of victim [7]. speed, checking driver license etc. Various approaches have
Intelligent Transport Systems (ITS) based on Internet of been used for this such as smartphones,VANETs, vibration
Things (IoT) are getting popular and can be seen as a solution sensors, pressure sensors, GPS, GSM and various machine
to improve the road safety. One effective technique to reduce learning algorithms. Even though there is literature available
traffic hazards and save precious lives could be to reduce on various strategies for accident detection and prevention;

VOLUME 8, 2020 122481

 U. Alvi et al.: Comprehensive Study on IoT Based Accident Detection Systems for Smart Vehicles

literature on accident detection and prevention techniques,

with the objective that smart systems can be developed with
improved accuracy and better strategies to control accident
causing factors while watching out for the existing challenges
in the current systems.
III. INTERNET OF THINGS CHARACTERISTICS
Internet of Things (IoT) refers to a network of things. These
things could be any object that we see in our daily life. These
objects are not limited to electronic devices or some high end
technology products but could include objects that we won’t
normally think of them as electronic like dustbins, chairs,
clothes etc., [21]. It’s a network of smart objects in which
the objects are capable of sharing resources and data with
other objects. The data collected from the objects is further
FIGURE 1. Real time accident detection using surveillance cameras. analyzed to extract the useful information [22]. The smart
objects are equipped with some sort of sensors, processing
however, no comprehensive survey exists. This paper aims powers and an ability to communicate to other objects. IoT
to fill this gap by critically reviewing the literature related to has provided many benefits to improve the quality of life,
accident detection, prevention, and reporting systems, to pro- where smart objects are used to simply our daily errands [23].
vide a broader perspective of existing techniques so that effec- The major characteristics of IoT are as follows:
tive systems can be developed that can utilize the strengths A. INTER CONNECTIVITY AND SERVICES
while addressing the challenges in the current systems.
As long as IoT is concerned, IoT devices can be inter-linked
The rest of this article is arranged as follows. Motivation
with the worldwide information and communication foun-
behind this work is discussed in Section II, Section III covers
dation [21], [24]. Things-related services such as privacy
the IoT characteristics, IoT applications in various domains
protection, semantic consistency between physical things and
are discussed in Section IV, Section V covers the typical chal-
their virtual interlinked items within the constraints of objects
lenges in an IoT environment, the emerging IoT technologies
are offered by IoT. Technologies continue to develop both in
are discussed in Section VI, Section VII covers the literature
the real world and in the information world to provide things-
review of various accident detection and prevention tech-
related services [21], [24].
niques, Section VIII presents a comparative study of different
techniques, Section IX presents recommendations based on B. HETEROGENEITY
the analysis and finally the paper is concluded in Section X. The IoT devices are heterogeneous in nature. Various devices
can communicate with each other due to this property. The
II. MOTIVATION BEHIND THIS STUDY devices that make use of different hardware platforms, net-
According to the World Health Organization (WHO) report, works, technologies can communicate with each other due
nearly 1.35 million people died in road accidents, making to this feature [21], [24]. IoT system includes connectivity
road traffic injuries the eight leading cause of death glob- of various devices, platforms and operating systems inter-
ally [20]. The number of fatalities associated with road acci- linked by different protocols [25]. IoT provides connectiv-
dents is extremely high, thus, measures must be taken to ity between different devices which have different power
improve road safety. Most injuries incurred by accidents are constraints, vendors and are designed for different purposes.
not serious, and the victim’s life can be saved if rescued The goal of IoT is to provide seamless machine to machine,
timely. However, it takes additional delay to manually notify machine to human and human to human connectivity, so it
the emergency teams due to poor communication mecha- must be able to connect different things and networks [26].
nisms, thus, victims are left unattended for a long time,
resulting in an increased death rate. C. DYNAMIC CHANGES
The consequences of road accidents are not just con- The IoT devices are dynamic i.e. rapidly changing in
strained to the loss of human lives yet, also incorporate the nature. The state of devices varies with time. For exam-
destruction of property, traffic blockages, and immense eco- ple, the devices may be in sleep/wake state, connected/
nomic loss. Thus automatic accident detection systems are the disconnected. The number of devices at a time can also
need of time, which can speed up the rescue operations and change [21], [24]. Some new devices can be added in the
limit the causalities after the mishap and numerous lives can network, while other might leave the network. IoT devices
be saved. This paper features existing mechanisms to detect are able to adapt to the changing needs. For instance, security
accidents, its working, and limitations. Furthermore, accident cameras can adapt their modes according to the time of the
prevention methodologies, accident contributing factors are day [25]. The device’s connectivity with other devices may
highlighted as well. This study critically reviews existing vary during different time intervals such that it’s connected

122482 VOLUME 8, 2020

U. Alvi et al.: Comprehensive Study on IoT Based Accident Detection Systems for Smart Vehicles

to one set of devices at one instance, and to another set at

another instance, so to ensure security, efficient cryptography
systems and good security protocols are needed [26].

D. ENORMOUS SCALE
The number of devices that are inter connected and need
to be tracked and handled would be at least an order of
magnitude greater than the internet devices currently con-
nected. It’ll become even more complex to manage the
massive amount of data produced by these devices and
using this data and it’s interpretation for different application
purposes [21], [24].

E. SAFETY
While we avail numerous benefits from IoT, an important
concern ‘safety’ should also be considered. Whether we are FIGURE 2. IoT applications.
creators or receivers of IoT, we must ensure safety. It includes
the safety of our physical well-being as well as our private B. IoT IN SMART ENVIRONMENT
data. This also includes ensuring the safety of communication
Weather monitoring – Climate changes such as humidity,
endpoints, networks, data being communicated across the
pressure, temperature and rain can be monitored.
paths, thus creating a security standard that will be able to
Water Quality Monitoring – Quality of water can be
scale [21], [24].
monitored, that is, whether it’s safe for humans to consume it.
F. CONNECTIVITY Prevention of Natural Disasters – River levels could be
Connectivity entitles attainability and compatibility. Network monitored for early flood detection, similarly land slides and
availability refers to be able to receive on the network and other natural disasters could be predicted.
compatibility refers to be able to use and produce data Air Quality Monitoring – Quality of air can be moni-
[21], [24]. New possibilities for IoT can be generated by tored and measures could be taken to control the emission
bringing together the routine objects and connecting the smart of CO2 and other toxic gases by vehicles, factories and
gadgets and applications [25]. industries.
Forest Fire Detection – Different fire conditions such as
IV. IoT APPLICATIONS combustion gases can be monitored to generate an early alert.
IoT has numerous and diverse applications which cover Protecting Wildlife – Wild animals can be traced and their
almost all areas in our day to day tasks. It covers many location can be determined through tracking collars,which
domains such as transportation, agriculture, healthcare, waste use GPS to track location and GSM for communication
management, supply chain, environment and energy as [21], [22], [24], [27].
shown in Figure 2. Some of the applications of IoT are as
follows: C. ROLE OF IoT IN INDUSTRIES
Explosive and Dangerous Gases – IoT can be utilized to
A. IoT IN HEALTHCARE detect gas leakages in manufacturing areas, in the vicinity of
IoT has numerous applications in healthcare. It can be used chemical plants and indoor mines. Harmful gases and oxygen
to monitor and indicate various health indicators. levels in chemical plants can be monitored and controlled to
Patient Monitoring – Patients can be monitored for var- ensure safety for both workers and products [21].
ious conditions like heart rate, blood pressure, glucose level Water, Oil and Gas levels Monitoring – The levels of
etc. inside hospitals. oxygen, water and oil can be monitored in storage tanks and
Medical Cold Storage – Various medicines, vaccines containers.
could be stored and their conditions can be monitored, like Maintenance and Repair – Early predictions can be made
when they’ll expire etc. for component malfunctions and maintenance service can be
Fall Detection – Assistance could be provided for set automatically before the actual component failure [21].
elderly or handicapped people if they fall, timely assistance Managing fleet of cars – The fleet of cars can be mon-
could be provided to them. itored for any corporation. It can monitor it’s performance
Dental – Toothbrush connected by means of Bluetooth to and can process the data to choose the one that needs
a Smartphone app can analyze brushing habits. maintenance [27].
Physical Activity Monitoring – Sensors placed on Temperature Monitoring – The temperature inside indus-
some gadget like watch, that a person could wear, it can tries, mines and other work places can be monitored to gen-
monitor how many steps he walked, heart rate etc., [21], [22], erate an alert, if the temperature exceeds to ensure the overall
[24], [27]. safety of workers and workplace.

VOLUME 8, 2020 122483

 U. Alvi et al.: Comprehensive Study on IoT Based Accident Detection Systems for Smart Vehicles

Ozone Detection – The presence of ozone levels could be trip, latest games and internet add the fun element to the
detected in food industries during the process of drying meat. trip [31].
Indoor Air Quality Surveillance – The quality of air can Collision Detection – In case of a vehicular mishap or acci-
be monitored inside workplaces to ensure general safety of dent, vibration sensors installed in the cars can detect the acci-
workers and goods [28]. dent. The information can then be communicated to provide
Rescue Operations – Factory workers who may have aid to the victims.
been stuck in underground units due to natural disasters Fire detection – The fire sensors installed in the vehicles
such as earthquake, land slides, explosion or some other can be used to detect the presence of flame or fire. On detec-
natural calamity may be saved, thorough deployment of IoT tion of fire the fire rescue system could be triggered or con-
resources that can accurately track their location [29]. cerned authorities can be notified.
Automated toll and fine payments – Every vehicle
D. IoT IN AGRICULTURE will have an RFID tag and on every toll booth, the sen-
IoT has vast applications in agriculture, some of them are as sors will scan the car and a predefined amount will be
follows: deducted from the owner’s bank account associated with that
Green House Monitoring – The climate conditions can automobile [32].
be monitored and controlled for green house to maximize Driving Insights – Different sensors like accelerometer,
the production of vegetables and fruits, and their quality Gyroscope, GPS etc. can be used to analyze the driving
can also be monitored by providing favorable environmental patterns of the driver.
conditions. Geo Fencing and Speed Monitoring – The system could
Animal Tracking – Animals grazing in open fields and indicate the driver if he has gone out of the bounds of prede-
pastures can be identified and located. fined geographical area or is over speeding by comparing it
Air Quality Monitoring – The quality of air can be moni- with the predefined threshold limit.
tored to detect any harmful or toxic gases that may emit from Driver Identification – Bio metrics can be used to authen-
waste material [21]. ticate the driver. Bio metrics may include, fingerprints, face
Field Monitoring – The condition of fields can be mon- recognition or voice. The driver voice could also be used
itored via different sensors, the data can then be processed, to provide voice commands to the navigation systems. Bio
and the farmer could be informed that a particular piece of metric data can also be used for anti-theft protection to ensure
land requires special care. maximum security [33].
Pest Control – Different mechanisms can be incorporated
to control the pests in crops, to ensure quality of crops [27]. V. IoT CHALLENGES
Water Management – The water can be managed effi- IoT improves the quality of our life, due to its numerous
ciently to minimize water wastage by making use of different applications. However due to it’s security and privacy issues,
sensors. it’s likely that it’ll be less adopted by the users [34]. IoT has
Soil Management – The condition of soil can be mon- vast uses and benefits in different sectors and solves many
itored such as measuring the PH levels, salinity, moisture problems, but still it has various challenges and limitations.
content etc. so that the farmer sow the seeds according to the One of the main challenge faced by IoT is maintaining privacy
soil level. and security of users’s data [35], [36]. IoT is a large scale
RFID tags and Sensors – Using RFID (Radio-Frequency network, which includes many manufacturers, industries, and
Identification) tags and sensors will help identifying and it may vary in different applications according to the user’s
recognizing the diseases that occurred in plants or crops. The need. Such large scale deployment of service, needs to be in
farmer can access the information from a remote location and the boundaries of a certain standard. IoT will be developed
can take the necessary actions, to save the crops [30]. in a step by step procedure. Various challenges faced by IoT
need to be addressed [37]. These challenges may include
E. IoT IN AUTOMOTIVE INDUSTRY power consumption, architecture challenge, heterogeneity,
Vehicle maintenance – The on board units installed on cars mobility, interoperability etc., [35]–[37]. Some of the chal-
can provide diagnosis of the car condition and can help the lenges of IoT are discussed below:
driver to find faults in the engine or some other parts. The
safety measures could also be ensured, like seat belt check, A. MOBILITY
drowsiness detection, over speeding etc. One of the major issue to implement IoT is mobility, because
Vehicle Tracking – The locations of vehicles can be IoT is expected to offer services to the mobile users as well.
tracked easily by simply installing IoT based trackers on the It continuously needs to connect the users, in order to provide
vehicles. better services. It is unable to provide the services to the
Entertainment with Connected cars – The connected mobile users when it needs to transfer from one gate to
cars become a center of Infotainment. It creates a pleasant another [35]. Mobility is one of the prominent characteristic
experience for the user. The apps on dash board can provide of IoT devices, thus the devices need to join the nearby
updates on traffic conditions, information about the current networks without any previous configuration. Thus, good

122484 VOLUME 8, 2020

U. Alvi et al.: Comprehensive Study on IoT Based Accident Detection Systems for Smart Vehicles

security mechanisms need to be implemented to make the IoT interoperability, syntactic interoperability, networking inter-
devices compatible with mobility [36]. operability, semantic interoperability and platform interoper-
ability. Many approaches have been proposed by researchers
B. POWER CONSUMPTION and industry to handle the interoperability issues, but still
Power drainage of devices is one important challenge in these approaches don’t cover all aspects of interoperability.
IoT. IoT is concerned with how it can interlink things in The collaboration between different vendors could also help
a compatible fashion, while watching out for the energy in solving the interoperability issues [42].
constraints because communication is one of the most power
consuming task [21]. Computing is involved in every aspect VI. PROMINENT IoT TECHNOLOGIES
of human lives, so power consumption becomes an unavoid- IoT consists of a large network of interconnected things
able issue. Some mechanisms should be introduced to have that communicate and exchange data. IoT is able to trans-
IoT devices which consume less energy. In order to use the form any real life object into a computing device that can
IoT devices, it’s need of time that problem of storing power sense and communicate [44]. It consists of a large network
in devices should be solved [38]. The energy capacity of IoT of heterogeneous devices consisting different sensors and
devices is limited and they have to be replaced after some actuators attached to various daily life objects [45]. IoT
time, however some devices consume a lot of power and they has turned the traditional internet where only human to
can’t be recharged. Low bandwidth connections should be human services were offered into a network where real life
used, in order to extend the battery life [36]. objects can communicate and exchange data [46]. Differ-
ent wireless and wired platforms can be used by the smart
C. SECURITY AND PRIVACY devices for communication purpose. Wireless IoT utilizes
To provide and avail services through out the day in an IoT different wireless protocols for communication [47]. These
environment, the things and people are connected with each technologies include LoRaWAN, Near Field Communication
other. However, the communication over internet is prone to (NFC), ZigBee, 6LoWPAN, Bluetooth Low Energy (BLE)
security breaches, since the devices are not equipped with and Z-Wave [44], [45], [47]. Some prominent wireless com-
good security mechanisms. Different devices are connected munication protocols used in IoT are discussed below:
with each other and data is shared among them instantly,
thus a mechanism is needed to ensure data integrity and A. LPWAN
confidentiality [39]. The network of IoT not only consists of Low Power Wide Area Network (LPWAN) is a wireless com-
things, but also crucial data and high value gadgets which munication technology. The main characteristics of LPWAN
further creates hurdles to ensure security. The main issues technologies are large communication range, long battery life
arise due to implementation through remote clouds which are and low cost devices [44]. In the LPWAN technology, there
connected with other interlinked systems, user’s personal data are two categories, namely, Long Range (LoRa) and narrow
etc., [40]. In order to make IoT more usable in real world band (NB-IoT) [48]. Lora is an unlicensed long range low
it’s security issues must be solved. The IoT network is esti- power wireless technology which improves network perfor-
mated to deal with a exponentially growing number of inter- mance, reduces device cost and supports large number of
connected devices. These objects will exchange information; devices [46]. NB-IoT is a licensed narrowband technology
thus their interactions must be secure to ensure data integrity. which provides improved performance in terms of range,
The heterogeneous nature of IoT, where different types of reliability, QoS (Quality of Service) and latency [49].
devices, located at different places will interact, further makes
it complex for the deployment of an efficient and scalable B. LoRaWAN
security algorithm [41]. Long Range Wide-Area Network (LoRaWAN) is a Low-
Power, Wide-Area Network (LPWAN) which has adequate
D. INTEROPERABILITY IN HETEROGENEOUS capacity and communication range with low power con-
ARCHITECTURE sumption and cost [44]. It offers core IoT requiremnts such
Interoperability is prominent challenge as far as IoT network as secure communication, mobility etc. It also reduces the
is concerned. This is mainly due to heterogeneous nature of complexity involved in communication due to heteroge-
the IoT devices which have varying protocols, data formats, neous nature of IoT by providing seamless interoperabil-
platforms etc., [42]. Internet connectivity requires that the ity [45]. It’s a wireless communication protocol developed
connected devices are able to communicate in the language by Lora Allience [46]. LoRaWAN standard supports two
understood by all of them, thus IoT systems need to han- security layers, to ensure application and network security.
dle the interoperability issues [21]. IoT platforms connect Device authentication is supported by network layer security,
various devices that may include different sensors, access whereas security of application data is ensured by applica-
points etc. Each platform may make use of different formats tion layer security [44]. LoRaWAN classifies end devices
for availing different services and resources. Therefore, it’s into three categories namely A, B and C. Class A devices
quite a challenge to provide services via such diverse plat- are in sleep mode most of the time and are energy effi-
forms [43]. IoT interoperability can be classified into device cient. LoRa gateway sends time synchronized beacons to

VOLUME 8, 2020 122485

 U. Alvi et al.: Comprehensive Study on IoT Based Accident Detection Systems for Smart Vehicles

Class B devices which enables them to open extra receive at a frequency of 13.56 MHz and a distance of about 10cm.
windows. Class C devices are able to receive data at any The devices in a NFC network communicate by generating
time except for the time when data is being transmitted [45]. fields [44]. NFC contains a tag which has data that can be
A typical LoRaWAN network has three types of entities, read only or the device can rewrite it later on. NFC operates
namely, Gateways (GWs), End Devices (EDs), and Network in three modes [47]. NFC technology makes it simpler to
Server (NS). LoRaWAN proved efficient in typical IoT appli- exchange digital content, making transactions and connecting
cations such as smart metering and environmental moni- electronic gadgets with a simple touch. It’s configuration and
toring [50]. LoRaWAN provides high flexibility, scalability, setup is easier to implement and it does not require line of
security and throughput [48]. sight [53].

C. NB-IoT VII. LITERATURE SURVEY

NB-IoT is a licensed narrowband technology and is time In this section different techniques for collision detection
slotted synchronous protocol [49]. It supports low complex- and reporting are discussed. Most of the systems focused
ity and low throughput Internet of Things. NB-IoT is built primarily on mechanisms for accident detection, while some
using existing LTE (Long Term Evolution) characteristics. systems also focused on taking preventive measures to pre-
Its network architecture is based on LTE network with little vent accidents, so the strategies used to detect and prevent
modifications to cater the needs of large network of users. accidents are discussed separately. Figure 3 summarizes the
NB-IoT requires a minimum bandwidth of 180 kHz to per- accident detection and prevention techniques discussed in this
form [51]. It has four downlink physical layer channels, study.
namely, broadcast channel, synchronization channel, control
channel and the data channel [52]. NB-IoT is preferred by
A. ACCIDENT DETECTION TECHNIQUES
applications that require high QoS and low latency [49].
Millions of people die in road accidents every year, most of
D. BLUETOOTH LOW ENERGY (BLE) the time accidents are not serious, but there is still a huge
Bluetooth Special Interest Group (SIG) developed Bluetooth loss of life due to delay of emergency services. So a system
Low Energy (BLE) technology in the 4.0 version of the Blue- capable of analyzing a situation and able to detect it as an
tooth. It’s also called Bluetooth Smart. It’s energy efficient accident would be very helpful in providing timely assistance.
and has low device costs [44]. It’s main characteristics include Various strategies used to detect accidents are discussed in
low energy consumption, easier setup and installation, and this section.
supporting star topology [47]. It’s protocol stack consists link
layer, physical layer and Logical Link Control and Adaptation 1) CONVENTIONAL ACCIDENT DETECTION TECHNIQUES
Protocol (L2CAP). The stack is classified into Controller a: USING VANET (VEHICULAR AD-HOC NETWORK)
and Host. The Controller is responsible to implement the Manuja et al. [10] proposed that problems of traffic conges-
link and physical layer, whereas the Host is responsible to tion arise due to vehicle failure or due to accidents in no
implement the upper layers [44]. The low energy consump- network area. They proposed a system to solve this problem
tion feature of BLE makes it a suitable protocol for many based on VANET. In this system every moving vehicle is
applications such as wearables, health monitoring apps, home considered as a node. The alert messages are transmitted
appliances etc. [45]. using RF module, and alert messages are received by the
moving vehicles that are in the range of RF module. Finally,
E. ZigBee the vehicle in the network area receives the message, then the
ZigBee is a wireless communication technology defined by received message is transmitted to the base station. The alert
the ZigBee Alliance. It provides seamless interoperability message contains four types of messages. These are detected
between devices from different vendors. The ZigBee protocol by piezo electric sensor, MEMS (Micro-electro-mechanical
stack is build up on IEEE 802.15.4 standard. To ensure secure systems) sensor, flame sensor, and temperature sensor.
communication it provides two types of keys namely, network The system identifies vehicular mishap in considerably less
key and multiple link key. The network key is shared by time and alerts the rescue teams along with the location of
all the nodes in network, whereas the multiple link key is mishap. A switch is also added in the system, which the
only shared among the communicating nodes [44]. ZigBee driver can use to stop sending alert message in case when
supports different types of topologies like tree, star and mesh injuries are minimum or there’s no serious loss. The con-
network topology. It’s suitable for those applications and troller gets the input from sensors and sends the accident
devices that require low data consumption, good battery life alert information to the road side unit and then message is
and maximum security [47]. sent to the rescue team. WIFI and GPS are used to find the
location of vehicle. In previous systems, GSM module failed
F. NFC to communicate in no network area, it’ll provide coverage
Near Field Communication (NFC) is a short-range communi- in network and no network area. A switch is also available
cation technology, which operates on electromagnetic fields to stop sending of message in case of no serious damage.

122486 VOLUME 8, 2020

U. Alvi et al.: Comprehensive Study on IoT Based Accident Detection Systems for Smart Vehicles

FIGURE 3. A taxonomy of accident detection and prevention techniques.

However, the major limitations that could be addressed in the services like rescue teams, ambulances who need to com-
future are VANETs have security and privacy issues, also they municate timely in case of emergency situations like road
have highly dynamic topology due to which routing issues accidents, mishaps, security alerts etc. The main applications
may arise. of VANETs include safety and non-safety applications. The
Shabir et al. [54] presented a survey of different conges- safety applications can reduce the chance of accidents by
tion control techniques for VANETs. To address the network providing early warnings to the vehicles. Time criticality is
congestion issue in VANETs three different approaches are one of the main concern in safety applications, however the
discussed which are proactive, reactive and hybrid congestion congestion in network can degrade it’s performance.
control strategies. To address network congestion, latency and A VANET is a type of MANET (Mobile ad hoc network)
throughput are two important parameters. The techniques are which considers vehicles as mobile nodes.VANETs can be
categorized in six major categories which include rate-based, used in ITS to ensure convenience and road safety. A typical
power based, hybrid strategies, CSMA/CA (Carrier Sense scenario for reporting emergency situations in a VANET
Multiple Access with Collision Avoidance) based, prioritiz- environment is demonstrated in Figure 4. The main issue
ing and clustering-based strategies. The discussions show in VANETs is, it’s highly dynamic topology due, to which
that proactive techniques are best to control congestion in different problems like network congestion, frequent dis-
VANETs since they proactively handle the congestion in net- connections etc. may arise. To address these issues, multi-
work, moreover priority based approaches also show better cast routing techniques can be used in VANET protocols.
performance if they are used along with hybrid congestion Different VANET multicast routing protocols are reviewed in
control techniques. this study. The performance of routing protocols is analyzed
In [55] its discussed that network congestion is a severe based on the routing techniques used by the protocols. The
issue in VANETs, since VANETs have highly dynamic topol- multicast routing techniques are classified as proactive, reac-
ogy. Congestion in VANETs can really affect the perfor- tive and flooding techniques. The goal behind development of
mance of applications using them, especially emergency multicast routing methods in VANETs is to address issues like

VOLUME 8, 2020 122487

 U. Alvi et al.: Comprehensive Study on IoT Based Accident Detection Systems for Smart Vehicles

continuously so in case of mishap the location is communi-

cated timely. The primary drawback of this approach is that,
no switch is available to cancel sending of alert messages in
case of no serious damage, secondly location is continuously
monitored, so it might drain a lot of battery.

d: USING ACCELEROMETER SENSORS

In [4] its discussed that,‘‘one fourth of the people involved
in accidents are motorcyclists’’. Rash and careless driving
are the primary causes of causalities in bike riders. A system
is proposed which uses helmet as an apparatus for accident
detection and reporting. Sensors, processors and cloud com-
puting infrastructure are used to build the system. On accident
detection details are sent to the emergency contacts using
FIGURE 4. Accident reporting scenario in VANET.
Cloud based service. GPS tracks the location of vehicle.
A GPS, microcontroller and an tri-axial accelerometer are
present on the helmet. The probability of accident is cal-
culated by continuously monitoring orientation of head and
network overhead, avoiding loops and adaptability in highly
helmet’s position. If the pre-fetched speed limit is exceeded,
dynamic topology. The VANET multicast routing protocols
a text message as an alert is sent to the emergency contacts.
are categorized into geocast and cluster-based routing. The
The system on the helmet connects with the cloud server
multicast routing protocols for VANETs are compared on
using RESTful architecture communicating through HTTP
various parameters such as real time, location based, map
(HyperText Transfer Protocol). The system is cost effective
based and spatiotemporal. The study shows that use of multi-
as compared to other systems, also on occurrence of accident,
casting in VANET routing protocols reduces fragmentation,
text messages are repeatedly initiated until acknowledged.
power consumption and transmission overhead. The setup for
The major drawback of this approach is that there is a possi-
multicast routing protocol for VANET may vary according to
bility of false alarm if the driver drops the helmet by mistake,
different environments, hence a single protocol may not be
additionally, more storage and computational capabilities are
suitable for every type of VANET environment, because the
needed because the system needs to communicate with the
design may change according to the needs and functionality
cloud services.
of that particular environment.
Anil et al. [58] presented a solution for collision detection.
The system detects the accident using flex and accelerom-
b: USING VIBRATION SENSORS eter sensors and the location of accident is communicated
Tushara and Vardhini [56] proposed a solution for accident via GSM modem to the emergency services. Alert message
detection that uses a micro controller to control operations contains information about location, vehicle number and time
like detecting and reporting. The system focuses on minimiz- of accident occurrence. The current situation of passengers
ing the action time after an accident has occurred. A number is is monitored through real time video transferred by the in
pre-fetched on the system, to which an alert is sent on occur- mounted camera. Accelerometer monitors speed of vehicle.
rence of accident. The accident is detected using a vibration Sensor’s output is compared with the pre-fetched value and
sensor. The system has high probability of generating wrong if it exceeds the defined threshold value, it indicates the
output because of its reliance on single sensor only. The occurrence of accident. The camera on the receiver module
weakness of the system is that it generates an alert message is connected with a screen, on which live situation of vehicle
on occurrence of accident but accident location is not shared. can be seen. The major advantage of this approach is that
the accident severity could be known and medical aid could
c: USING PIEZOELECTRIC SENSORS be provided accordingly due to in mounted cameras in the
Patil et al. [57] put forward a solution in which the system vehicle. However, this also has privacy concerns due to live
will sense accidents and will inform the nearest police station video feed transmission.
and rescue teams. GSM technology is used to communi- Megalingam et al. [59] proposed a solution to smartly
cate alert messages to the emergency services. The system detect vehicle collision and alert the rescue teams. The system
continuously tracks the vehicle and can immediately alert can be placed inside a vehicle in which an accelerometer
the emergency services in case of any mishap. Renessa’s sensor detects an accident, this output of sensor is monitored
microcontroller is used in conjunction with GSM modem by a microcontroller. The RF (Radio Frequency) transmitter
and GPS receiver. GSM is used for communication purpose module is interfaced with the microcontroller which transmits
which sends an alert message containing the location which the information to the emergency services. The RF receiver
is provided by GPS. It’s main modules consist of piezoelec- module at the emergency service will receive this informa-
tric sensor, GSM and GPS. The system tracks the location tion. The emergency services on receiving the information

122488 VOLUME 8, 2020

U. Alvi et al.: Comprehensive Study on IoT Based Accident Detection Systems for Smart Vehicles

would take appropriate actions to provide timely medical aid. lights sense RF signal, they turn green to allow the ambulance
The major drawback of this approach is that RF module has reach the accident spot as fast as it could. The emergency
limited range, moreover it relies only on one sensor, so it has message is also sent to the victim’s family.
single point of failure.
g: USING SHOCK SENSORS, GPS
e: USING PRESSURE SENSOR, TILT ANGLE Nasr et al. [61] presented a solution in which the system
In [12] a solution based on android application for accident informs the PSO (Public Safety Organization) on occurrence
detection and reporting is presented. Outward force experi- of accident. Shock Sensor detects occurrence of accident,
enced by the vehicle is monitored by means of an external the signal is processed and then sends the geographic loca-
pressure sensor, speed and tilt angles are measured by GPS tion to PSO. The system has different phases which are as
and accelerometer sensors of the phone. Bluetooth on the follows:
phone receives the data from sensors. As accident occurs (a) Registration of vehicle
the speed of vehicle decreases abruptly. So an accident will (b) Registration of passenger
be detected if vehicle speed changes abruptly and values of (c) Accident monitoring through PSO headquarters
pressure and tilt angle exceed the pre-fetched threshold limit. In Vehicle Registration phase, the person’s vehicle is reg-
A switch which the driver can use to stop sending of alert mes- istered. An IoT device is installed on the system. After instal-
sage in case when the accident isn’t severe or in case of false lation of device, the person gives the vehicle ID to the person
alarm is also present. A smartphone based accident detection who’s responsible for registration of vehicles in the PSO
system is demonstrated in Figure 5. However, the research headquarters’ database. Its main modules consist of NFC
is subject to some limitations such as magnitude of force reader, shock sensor, GPS, and cellular IoT. These modules
experienced by the smartphone would not be the same as of communicate with each other to identify an accident and
the magnitude of force experienced by the vehicle, moreover report it to the PSO headquarters with the exact location
smartphones have battery constraints. of accident. As the shock sensor detects accident, an alert
through HTTP request is sent along with the location.

h: USING GPS, GSM

In [5] a system is proposed, that considers the speed as one
of the major cause of accident. It uses a GPS receiver to
monitor the speed and detects an accident based on monitored
speed. The GPS module continuously monitors the speed and
compares with the previously monitored speed every second
using a micro-controller unit.
Whenever the system identifies that the speed is less than
the pre-fetched threshold limit it will detect the situation as
an accident. The location is detected by using a GPS module.
An alert message is sent to the emergency services using a
FIGURE 5. Smartphone based accident detection system.
GSM module. Figure 6 illustrates a GPS/GSM based accident
detection system. The problem of the system is that it relies
only on a single sensor for accident detection which has high
f: USING GYROSCOPE, GPS, ANDROID APP chance of generating false positives.
AND RF TRANSMITTER
Kumar and Deepak [60] proposed a solution for accident
detection and reporting which requires no manual interaction
of human before or after the occurrence of accident. Its main
module consists of GPS, Raspberry Pi, android application
and Gyroscope. The system also stores the blood group of
the driver. As soon as the vehicle meets an accident, the gyro
which is placed in the vehicle measures the rotation angle
and if it exceeds the threshold limit, it will send an alert
to the active server. The app is placed in the ambulance
which continuously receives the information from server,
the app helps the ambulance to navigate to the accident spot FIGURE 6. An accident detection system using GPS and GSM.
using the route which has less traffic by using Google maps
API (Application Programming Interface). To avoid traffic In [62] a solution to find the position of vehicle based
congestion, the ambulance transmits RF signal time to time, on ARM micro controller is presented. To get the position
the RF receiver is present in the traffic lights. When the traffic of vehicle, the intended person initiates a request via SMS.

VOLUME 8, 2020 122489

 U. Alvi et al.: Comprehensive Study on IoT Based Accident Detection Systems for Smart Vehicles

The request is received by the GSM module and is acknowl- obtained from the authorities, for the road that had increas-
edged. The request is then processed by the Spartan proces- ing accidents in the past years, to improve the accuracy
sor. The processor sends the command to GPS module of the of the model. The model maps the crisp inputs to corre-
system. The GPS module tracks the location of vehicle and sponding crisp outputs using fuzzy rules based on conditions.
replies the request with location coordinates of the vehicle. Every input variable is defined by the membership functions.
Vehicle’s location is sent to the user with longitude and In fuzzification process the fuzzy variables are constructed
latitude’s values. The weakness in the literature lies in the fact by deriving membership function for inputs and outputs
that external environmental factors can have greater impact and linguistic representation of these membership functions.
on the performance of system. Triangular and trapezoidal functions are used in this model.
The results of the model show good accuracy in predicting
2) ML/AI BASED ACCIDENT DETECTION TECHNIQUES accidents. The factors like road lighting and weather condi-
Transportation system plays an important role in human life; tions were ignored in this study.
however, where it provides many facilities, it has many risks
associated with it as well [63]. The road traffic accidents are b: ACCIDENT PREDICTION USING SUPPORT
increasing every year, so an effective mechanism is needed VECTOR MACHINE (SVM)
to minimize its frequency [16]. Crash prediction models B. Pan and H. Wu [66] proposed an approach to detect acci-
have been very famous to ensure road safety particularly dents by means of mobile sensors and SVM. Their approach
on highways [64]. Different models have been presented particularly focused on detecting accidents occurring in urban
for crash prediction in the domain of Machine Learning roads, since they are more prone to accidents, due to involve-
(ML) and Artificial Intelligence (AI), namely, Artificial Neu- ment of many flow disruptive entities like bus stops, traffic
ral Networks (ANN) [65], Support Vector Machine (SVM) signals etc. as compared to freeways where the traffic flow is
[64], [66], fuzzy logic [19], [63], Genetic Programming [67], not disturbed. This approach uses VANETs in which every
Random Forest Classifiers [16] etc. The real time data is vehicle is able to collect its own traffic data like location,
fetched in the models and is compared with the previously identity, speed, lane state with the help of mounted sensors
collected accident data, which helps in differentiating a nor- on vehicles. The traffic data is then collected by RSU (Road
mal situation from an accident [17]. The key approaches Side Unit) for further processing which is sent by On Board
for accident prediction in the ML/AI domain are discussed Units. Three traffic variables are considered to detect accident
below: which are vehicle’s speed, acceleration and lane-changing
state. Wider range of monitoring can be provided by means
a: ACCIDENT DETECTION USING FUZZY LOGIC of on-board mobile sensors.
Alkandari et al. [19] proposed a solution to detect accidents
in range of traffic lights using fuzzy logic technique which c: ACCIDENT DETECTION USING ARTIFICIAL
detects accidents. The system consists of two sub systems NEURAL NETWORK
namely, Detection System and Action System. The system In [18] a system is proposed that can detect an accident from
is based on Webster Method with a little variation. The the video footage obtained from the CCTV cameras installed
system collects data about different zones including number on highways. Each frame of video is given as an input to
of cars in a lane, speed of cars in particular lanes etc. The the Convolutional Neural Network (CNN) model, which is
disturbance in normal traffic flow is the main indicator that able to distinguish frames of video into accident and non-
an accident has occurred. The main elements of the system accident. Raspberry Pi 3 B+ Model is used as a remote
are crisp inputs/outputs, membership functions, fuzzy rules computer that can be placed on the CCTV cameras. A Pi
and linguistic variables. The linguistic variables which are, Camera is used in the system for the purpose of demonstration
cross ratio, zone status, accident status and section speed to obtain the video data. Inception v3 model is used to detect
help in determining a situation as an accident. The output accidents by prior training it, on two different set of video data
of the system is generated by applying the fuzzy rules to consisting of accident and non-accident frames. The model
the linguistic inputs, depending upon the input and rules a is based on CNN which is useful for image classification,
suitable action would be taken to improve the traffic flow. object detection etc. The proposed model uses both CNN
The results show that the system is able to identify most of and Long Short-Term Memory (LSTM) layers. The inception
the accident scenarios with good accuracy. v3 can work on heterogeneous convolutions which enables
In [63] an accident prediction model using fuzzy logic it to extract more features from the images. The model is
is presented. The relation between accidents and the fac- implemented on Raspberry Pi by using Tensor Flow, Open
tors contributing it are non-linear thus using fuzzy logic is CV and Keras. Each video frame is run through the model
a preferable choice for non-linear relationships. Many fac- and then prediction is given whether the frame is accident
tors like traffic conditions, human negligence etc. contribute frame or not. A threshold limit is set and if the prediction limit
towards road accidents. The major inputs to the model are exceeds 60%, the GSM module is triggered by the Raspberry
number of vehicles passing in a lane, speed, road width and Pi, which sends an alert message to the nearest hospitals
road surface condition. Accident data used in the model was along with the location coordinates. The results show good

122490 VOLUME 8, 2020

U. Alvi et al.: Comprehensive Study on IoT Based Accident Detection Systems for Smart Vehicles

accuracy of the system for the trained set of data, moreover

the system is also cost effective. The limitation of the study
is that, the model was trained for severe accident conditions
data, thus the model may not cover all the accident scenarios
where accident may be moderate or minor.

B. ACCIDENT PREVENTION TECHNIQUES

Accident prevention can be defined as the strategy or an
approach, or actions taken to avoid or stop an accident before
it occurs. Majority of the accidents occur due to human
negligence. These factors are over speeding, traffic law vio-
lations, drunk driving etc. So controlling these factors can
help to avoid accidents and save the precious lives. Different
approaches used to prevent accidents are discussed in this
section.

1) CONVENTIONAL ACCIDENT PREVENTION TECHNIQUES

a: DRIVER DROWSINESS CHECK FIGURE 7. Workflow of alcohol sensor based systems to prevent drunk
driving.
Eduku et al. [68] presented a solution to prevent traffic haz-
ards by proposing a system which uses eye blink sensor and
automatic braking system to slow down the car and bring it is not drunk, moreover external interference of air can also
in the state of halt, if drowsiness is detected in driver. A RF affect the performance of alcohol sensor.
module routs the information to the nearby vehicles to alert
them that a halt car is there. IR sensors are used to monitor the c: MAINTAINING SAFE DISTANCE USING
eye blink and detect the state of drowsiness. Infra-red rays are ULTRASONIC SENSORS
transmitted to the eyes by means of a IR transmitter, the IR
Kasera [69] presented a system to prevent accidents while
receiver on the other hand receives the reflected rays from the
driving in hilly areas. Since the roads on the hilly areas are
eye. If the output of IR receiver tends to be high, it implies that
very steep and curvy. To implement this mechanism, the sys-
the eyes are closed. Thus the drowsiness is detected by means
tem uses ultrasonic sensors. The system alerts the driver going
of monitoring this high/low output. Its main modules consist
on one side of the road about the vehicle coming from the
of an alarm that warns the driver if drowsiness is detected,
opposite direction of the road. The ultrasonic sensors are
automatic breaking systems to slow down the car, and bring
placed along one side of the road before curve and a LED
it to halt state and RF module to send alert messages to the
light after the curve, so if the vehicle comes from one end of
nearby vehicles in range. The main limitation of the system
the curve the sensor will detect it and LED light lights up on
is that the range of RF module is low.
the opposite side of the road. Two ultrasonic sensors and two
LED lights are installed on both sides of the curve. As soon
b: USING ALCOHOL SENSORS, GPS, GSM as the vehicle reaches one end of the curve, the RED light on
Rao and Yellu [3] proposed a system to prevent accidents the opposite side will automatically light up and will remain
using alcohol sensors to prevent drunken driving, since 70% on, unless and until the sensor on the other side of the curve
of the cases of accidents are due to ‘‘drunk and drive’’. detects it. It’s a low power consumption system, moreover its
An alcohol sensor placed in the system, is used to monitor implementation is very cost effective.
the contents of alcohol in blood. A workflow of such a system
is illustrated in Figure 7. Sensor is placed above the steering 2) ML/AI BASED ACCIDENT PREVENTION TECHNIQUES
so that, as the driver breathes the sensor could determine the a: DRIVER DISTRACTION IDENTIFICATION WITH DEEP
alcohol level. A threshold is set and if the contents are more CONVOLUTIONAL NEURAL NETWORK (CNN)
than the pre fetched limit, the car doesn’t move. GPS detects In [70] it’s discussed that, distracted driving is leading cause
the location of the vehicle. Messages are sent by means of of road collisions. Distracted driving can be any activity
GSM to the pre-fetched mobile numbers and inform them that which makes the driver lose his attention and focus while
the person is highly intoxicated and can’t drive. The system driving. Some of these activities include talking on phone,
focuses on the preventive measures to avoid any hazardous texting, adjusting the radio etc. The proposed technique iden-
situation. The system would be helpful to public in general as tifies distracted driving using CNN. Driver distraction can
well, since it won’t allow drunk person to drive. However, its be classified into different categories namely, visual, manual
main limitation is that there could be a false alarm in case and cognitive. Multiple types of distraction can exist together
when the person sitting next to driver is drunk but the driver in an activity. Distracted drivers can be identified using

VOLUME 8, 2020 122491

 U. Alvi et al.: Comprehensive Study on IoT Based Accident Detection Systems for Smart Vehicles

various measures, this technique particularly focuses on using time zones namely, morning, noon, and evening time. After
visual data to identify distracted driver. A camera is installed the merging algorithm terminates, final detection list contains
inside the car cabin to identify the driver’s behavior. Deep the final results of the detection system. If the height and
learning model is used to classify distracted driving from position are within the margin of tolerance, the result was
normal driving. In this study, a comparison of pre-trained considered correct. This approach achieves high accuracy
deep CNN is presented. The data set is divided into 10 distinct through automatic optimization of the features and regular-
classes to identify different distraction behaviors like texting, ization of neural network. The accuracy of the classifier is
drinking, talking to other passenger etc. The comparison improved by using Haar wavelet features.
shows that VGG16 with S2 shows good performance in terms
of accuracy. The models were pre-trained on the Image net C. HYBRID TECHNIQUES
data set, and then retrained on State Farm data set to identify Road collisions are one of the leading cause of fatalities.
distracted driving behaviours. In most cases, injuries are not serious and if victims are
rescued in time, lives may be saved. Many factors contribute
b: DRIVER DROWSINESS DETECTION USING to accidents, including driver negligence, drowsy and drunk
REPRESENTATION LEARNING driving etc. Thus a system that can detect as well as con-
Dwivedi et al. [71] proposed a methodology to detect drowsy trols the factors contributing to accidents will be helpful in
state of a driver using representation learning to prevent acci- preventing accidents and saving lives. Hybrid techniques are
dents caused by sleep. The model extracts the visual features techniques which use both accident detection and prevention
from the data, the features are learnt by using CNN. The mechanisms, as shown in Figure 8. Some of the systems based
input image is combined with the learned weights to produce on hybrid techniques are discussed below.
feature maps. A soft-max classifier layer uses these set of fea-
tures to classify the frames into drowsy and non-drowsy. The
facial features are derived using CNN based representation
learning technique by which the complex relationships of raw
data can be represented by combining features of features.
The model consists of two convolutional layers along with
a hidden layer of sigmoid function which is linked with
regression layer for classification purpose. The frames are
extracted from the video and are fed to a face detector. The
detected faces are then cropped to square images and then the
images are normalized, which are then fed to a convolutional
neural network. The classifier shows good performance on
diverse data set. The major advantage of the approach is that,
FIGURE 8. An accident detection and prevention system.
representation learning can capture more intelligent features
from raw input data as compared to manual methods.

c: PEDESTRIAN DETECTION WITH CNN 1) USING LIMIT SWITCHES

A considerable amount of pedestrians die every year in road Berade et al. [11] put forward a solution for vehicular mishap
crashes. Szarvas et al. [72] proposed the use of CNN as a detection by means of GPS to determine the mishap loca-
classifier in a pedestrian detection system, to identify the tion and GSM to communicate the information to the pre-
pedestrians that could help in reducing the number of acci- fetched number. Additionally, the system also identifies if
dents. The algorithm is tested on pedestrians in an urban envi- the driver isn’t wearing seat belt or if he is intoxicated. The
ronment. CNN classifier is compared with the SVM (Support sensor detects the accident and then PIC (Programmable
Vector Machine) classifier and results show that CNN wins Interface Controllers) microcontrollers send signals to GPS,
over SVM in terms of accuracy using classic image features. GPS tracks the location and then the alert message is sent
The candidates’ images are given as an input to the classifier. through GPS module. Limit switches are used for accident
A classification score is given to each candidate by the CNN, detection. Seat belt test will ensure that the driver is wearing
candidates with higher scores than the threshold are saved in seat belt and alcohol sensor placed on steering will check if
the raw detection list. The list is then sorted by classification driver is drunk. The accident is detected via limit switches
scores. placed at the back and front of the car. A LCD screen displays
Multiple detection merging operation is used to remove messages, in case of seat belt is not worn, or if driver has
redundant results for the same pedestrian. The evaluation data consumed alcohol, LCD screen gives warnings. System not
set consists of positive samples (pedestrians) and negative only focuses on accident detection but also on its prevention
samples (background) obtained from the camera mounted by introducing seat belt and alcohol sensors. The use of
on the top of car. In order to consider variety of lighting PIC micro controllers in this system is very effective since
conditions, the data set was equally distributed in different they consume very less power as compared to other micro

122492 VOLUME 8, 2020

U. Alvi et al.: Comprehensive Study on IoT Based Accident Detection Systems for Smart Vehicles

controllers and are also very fast. The findings of this study by controlling speed when a critical point is reached. Email
have to be seen in light of some limitations, as no switch is alerts are sent in case of accident, which are slow means
included to cut off the signal to the rescue teams in case the of communication, and providing medical aid could take
driver is safe. time.

2) DROWSY AND DRUNK DRIVING DETECTION 4) USING CELL PHONE,SPEED AND

Nanda et al. [2] presented a system which detects accidents ACCELEROMETER SENSORS
by means of vibration sensors. The solution kept in mind Ahmed and Jawarkar [13] presented a solution for accident
the needs of bike riders as they are more prone to serious detection and prevention. A system based on microcontroller
harm as bikes aren’t equipped with preventive measures as and an accelerometer sensor along with a recycled cell phone.
compared to other vehicles. GPS tracks the location and The system warns the user in case of unsafe driving and
for communication purpose GSM is used. The system also sends alerts in case of abnormal conditions of engine temper-
has a mechanism for checking if the driver is sleepy. The ature or accelerometer. It also has a feature to capture image
system also checks if the rider has driving license. The system after accident occurrence and send the multimedia messages
also makes sure that the driver hasn’t consumed alcohol. to pre-fetched numbers. Its main components include speed
The camera placed on the front of bike monitors the traffic sensor, accelerometer sensor, cell phone interface, tempera-
signals and automatically lessens the speed. It makes use of ture sensor and Real Time Calendar (RTC) chip. It’s environ-
many measures like traffic signal detection, checking driver ment friendly since junk cell phones are recycled. It can be
license and checking the state of driver to prevent accidents. easily implemented on existing vehicles. The major limitation
The system not only detects but also focuses on preventive of this approach is that on occurrence of accidents, alert
measures to avoid accidents thus, saving precious lives. The messages have to be sent manually by the user.
major limitation of this approach is, an inexperienced person
can use another person’s license to drive, since the system 5) SMART HELMET
does not authenticate if the license belongs to the person In [14] a system is presented, which uses smart helmet as an
driving. apparatus for accident detection and prevention. The system
won’t allow the rider to drive unless and until he wears a
helmet. The pressure sensor and accelerometer sensor are
used to check if helmet is worn by the rider or not. The alcohol
sensors placed on the helmet check the alcohol content in
the drivers breathe. If the contents are more than the pre-
defined threshold limit, the bike won’t start, thus preventing
drunken driving. On occurrence of crash the helmet worn
by the rider will hit the ground,the sensors will detect the
motion and tilts of helmet,and will immediately send the site
of crash to the emergency contact numbers and rescue teams.
The system would be very helpful to the driver as well as the
general public, since drunk driving not only risks the life of
driver but also affects the safety of other people driving on the
roads.
FIGURE 9. Obstacle distance based collision prevention system.
VIII. COMPARATIVE ANALYSIS
In this section, various systems discussed in the literature
3) USING DISTANCE SENSORS survey are compared according to the parameters defined
In [73] a system is proposed that always monitors the dis- in Table 1. The systems are compared in terms of accuracy,
tance between vehicles and obstacles that are in front, using their implementation cost to incorporate them in existing
distance sensors. A system maintaining safe distance from vehicles, what’s their communication mechanism to send
obstacles is demonstrated in Figure 9. The proposed system alert messages to the rescue and emergency teams, are
alerts the driver to control the speed and reduce the speed by the messages sent automatically on detection of an acci-
itself when a critical distance comes. Whenever an accident dent or they have to be sent manually by the user etc. Since no
takes place for uncertain condition, an email alert will be sent system is perfectly accurate, there’s always a chance of error
to the accountable individual with car details. The system in any system, so the systems are also compared against their
automatically controls the speed in case when critical point is chance of generating a false positive, i.e. triggering of alarm
reached. Vehicle’s distance is measured by ultrasonic sensors. in non-accident circumstances.
The system makes use of Raspberry Pi, Ultrasonic Sensor, Moreover, in some cases, the accident severity may not be
Led Buzzer and Servo motor. On occurrence of accident the serious and the driver may not need immediate medical aid,
alert message is sent using Raspberry Pi. It prevents accidents so in those cases a mechanism should be there to stop the

VOLUME 8, 2020 122493

 U. Alvi et al.: Comprehensive Study on IoT Based Accident Detection Systems for Smart Vehicles

TABLE 1. Comparative analysis.

sending of an alert message, to save the time of rescue teams. IX. RECOMMENDATIONS
So the presented systems are also compared if they’ve the Various methods for accident detection and prevention were
mechanism to stop sending of an alert message or not. discussed in this paper. The methods included warning the
It can be seen from Table.1 that many systems lack accu- driver for over-speeding, maintaining safe distance from other
racy, and may not generate accurate results in every situation. vehicles, avoiding intoxicated and distracted driving etc.
Moreover, many systems did not have a mechanism to stop Integration of these systems with the vehicles would be very
sending of alert messages in case of false alarm. beneficial to the society. These systems would be effective
Most of the systems presented in the literature survey, in minimizing the casualties associated with road accidents.
relied on some sort of hardware based technologies like Additionally, patient history such as blood group, age, aller-
sensors for implementing accident detection and preven- gies etc. can also be included in these systems to provide med-
tion mechanisms. A summary of different type of sensors ical aid accordingly. Moreover, data obtained from sensors
such as alcohol, seat belt, vibration, pressure etc. used in after an accident has occurred, can be used in data mining to
these systems, for accident detection and prevention is pre- deduce important results. Performing analysis on the data can
sented in Table 2. It’s quiet evident from the analysis that, give us valuable insights on how most of the road accidents
most of the systems use GPS to detect location of acci- occur, which factors contribute the most in event of mishaps,
dent, GSM as a communication mechanism to send alert which roads are dangerous and the time stamp in which
messages and accelerometer and pressure sensors to detect most of the accidents occur. The data collected from these
accidents. systems could also help police to find the crimes of hit and

122494 VOLUME 8, 2020

U. Alvi et al.: Comprehensive Study on IoT Based Accident Detection Systems for Smart Vehicles

TABLE 2. Summary of sensor types used for conventional accident detection and prevention techniques.

run cases. In addition, the information about occurrence of that those sensors or devices can themselves be destroyed in
accidents can be routed to the vehicles in range, to avoid any the accident and can generate erroneous readings and results.
further mishaps, that can be helpful in reducing chain reaction So such frameworks are required which are less reliant on
accidents in which multiple vehicles are involved in the crash, some kind of hardware or software.
creating a chaotic situation.
REFERENCES
[1] F. Bhatti, M. A. Shah, C. Maple, and S. U. Islam, ‘‘A novel Internet of
X. CONCLUSION AND FUTURE WORK Things-enabled accident detection and reporting system for smart city
The number of casualties associated with road collisions is environments,’’ Sensors, vol. 19, no. 9, p. 2071, May 2019.
growing rapidly. If victims are rescued in due time, several [2] S. Nanda, H. Joshi, and S. Khairnar, ‘‘An IOT based smart system for acci-
dent prevention and detection,’’ in Proc. 4th Int. Conf. Comput. Commun.
lives may be saved. We discussed various strategies which Control Autom. (ICCUBEA), Aug. 2018, pp. 1–6.
focused not only on accident detection but also on its pre- [3] T. V. N. Rao and K. R. Yellu, ‘‘Preventing drunken driving accidents using
vention. These strategies utilized various sensors such as IoT,’’ Int. J. Adv. Res. Comput. Sci., vol. 8, no. 3, pp. 397–400, 2017.
[4] S. Chandran, S. Chandrasekar, and N. E. Elizabeth, ‘‘Konnect: An Internet
accelerometer sensors, shock sensors, pressure sensors etc. of Things(IoT) based smart helmet for accident detection and notification,’’
and various machine learning techniques such as neural net- in Proc. IEEE Annu. India Conf. (INDICON), Dec. 2016, pp. 1–4.
works, support vector machines, representation learning etc. [5] M. Syedul Amin, J. Jalil, and M. B. I. Reaz, ‘‘Accident detection and
reporting system using GPS, GPRS and GSM technology,’’ in Proc. Int.
for accident detection. Various strategies for accident preven- Conf. Informat., Electron. Vis. (ICIEV), May 2012, pp. 640–643.
tion were also addressed, which include detection of drunk [6] J. B. Edwards, ‘‘The relationship between road accident severity and
and drowsy driver, regulating vehicle speed, maintaining safe recorded weather,’’ J. Saf. Res., vol. 29, no. 4, pp. 249–262, Dec. 1998.
[7] R. N. Nayan Kumar, V. R. Navya, N. Ganashree, K. U. Pranav, and N. Ajay,
distance from obstacles etc. Once the accident is detected, ‘‘Intelligent vehicle accident detection and ambulance rescue system,’’ Int.
the information is communicated to emergency services to J. Advance Res., Ideas Innov. Technol., vol. 5, no. 3, pp. 685–687, 2019.
provide timely aid. Such systems provide many advantages [8] M. U. Ghazi, M. A. Khan Khattak, B. Shabir, A. W. Malik, and
M. S. Ramzan, ‘‘Emergency message dissemination in vehicular net-
such as mitigating road collisions, identifying precise acci-
works: A review,’’ IEEE Access, vol. 8, pp. 38606–38621, 2020.
dent locations and facilitating over all rescue operations. The [9] W. Farooq, M. Ali Khan, and S. Rehman, ‘‘A novel real time framework
integration of these systems with vehicles would be somehow for cluster based multicast communication in vehicular ad hoc networks,’’
expensive yet will give various advantages. However, the sys- Int. J. Distrib. Sensor Netw., vol. 12, no. 4, Apr. 2016, Art. no. 8064908.
[10] Manuja M, Kowshika S, Narmatha S, and G. Theresa W, ‘‘IoT based
tems we discussed were all reliant on some kind of hard- automatic accident detection and rescue management in Vanet,’’ SSRG Int.
ware or software based technology and there is a possibility J. Comput. Sci. Eng., pp. 36–41, Feb. 2019.

VOLUME 8, 2020 122495

 U. Alvi et al.: Comprehensive Study on IoT Based Accident Detection Systems for Smart Vehicles

[11] P. Berade, K. Patil, P. Tawate, and M. U. Ghewari, ‘‘Intelligent accident [35] O. Helmi, M. A. Sokeh, and G. Sepidnam, ‘‘The challenges facing with the
identification and prevention system using GPS and GSM modem,’’ Int. Internet of Things,’’ Int. J. Sci. Study, vol. 5, no. 4, pp. 527–533, 2017.
Res. J. Eng. Technol., vol. 5, no. 2, pp. 1915–1919, 2018. [36] S. I. Al-Sharekh and K. H. A. Al-Shqeerat, ‘‘Security challenges and
[12] A. B. Faiz, A. Imteaj, and M. Chowdhury, ‘‘Smart vehicle accident detec- limitations in IoT environments,’’ Int. J. Comput. Sci. Netw. Secur., vol. 19,
tion and alarming system using a smartphone,’’ in Proc. Int. Conf. Comput. no. 2, pp. 193–199, 2019.
Inf. Eng. (ICCIE), Rajshahi, Bangladesh, Nov. 2015, pp. 66–69. [37] S. Chen, H. Xu, D. Liu, B. Hu, and H. Wang, ‘‘A vision of IoT: Applica-
[13] V. Ahmed and N. P. Jawarkar, ‘‘Design of low cost versatile microcontroller tions, challenges, and opportunities with China perspective,’’ IEEE Inter-
based system using cell phone for accident detection and prevention,’’ in net Things J., vol. 1, no. 4, pp. 349–359, Aug. 2014.
Proc. 6th Int. Conf. Emerg. Trends Eng. Technol., Nagpur, India, Dec. 2013, [38] P. Matta and B. Pant, ‘‘Internet-of-Things: Genesis, challenges and appli-
pp. 73–77. cations,’’ J. Eng. Sci. Technol., vol. 14, no. 3, pp. 1717–1750, 2019.
[14] C. Priyanka, P. Darshini C, V. G. Shavi, and S. Begum, ‘‘Two-wheeler [39] M. Abdur, S. Habib, M. Ali, and S. Ullah, ‘‘Security issues in the Internet
safety system for accident prevention, detection and reporting,’’ Int. J. Eng. of Things (IoT): A comprehensive study,’’ Int. J. Adv. Comput. Sci. Appl.,
Comput. Sci., vol. 7, no. 03, pp. 23680–23682, Mar. 2018. vol. 8, no. 6, pp. 383–388, 2017.
[15] M. Ozbayoglu, G. Kucukayan, and E. Dogdu, ‘‘A real-time autonomous [40] Mandal K and G. S. P. Ghantasala, ‘‘A complete survey on technological
highway accident detection model based on big data processing and com- challenges of iot in security and privacy,’’ Int. J. Recent Technol. Eng.,
putational intelligence,’’ in Proc. IEEE Int. Conf. Big Data (Big Data), vol. 7, no. 6S4, pp. 332–334, 2019.
Dec. 2016, pp. 1807–1813. [41] R. Roman, J. Zhou, and J. Lopez, ‘‘On the features and challenges of
[16] N. Dogru and A. Subasi, ‘‘Traffic accident detection using random forest security and privacy in distributed Internet of Things,’’ Comput. Netw.,
classifier,’’ in Proc. 15th Learn. Technol. Conf. (L T), Feb. 2018, pp. 40–45. vol. 57, no. 10, pp. 2266–2279, Jul. 2013.
[17] P. Amrith, E. Umamaheswari, R. U. Anitha, D. Mani, and D. M. Ajay, [42] S. Jabbar, F. Ullah, S. Khalid, M. Khan, and K. Han, ‘‘Semantic inter-
‘‘Smart detection of vehicle accidents using object identification sensors operability in heterogeneous IoT infrastructure for healthcare,’’ Wireless
with artificial intelligent systems,’’ Int. J. Recent Technol. Eng., vol. 7, Commun. Mobile Comput., vol. 2017, pp. 1–10, Mar. 2017.
no. 5, pp. 375–379, 2019.
[43] J. Koo, S.-R. Oh, and Y.-G. Kim, ‘‘Device identification interoperability in
[18] S. Ghosh, S. J. Sunny, and R. Roney, ‘‘Accident detection using convolu- heterogeneous IoT platforms,’’ Sensors, vol. 19, no. 6, p. 1433, Mar. 2019.
tional neural networks,’’ in Proc. Int. Conf. Data Sci. Commun. (IconDSC),
[44] D. Dragomir, L. Gheorghe, S. Costea, and A. Radovici, ‘‘A survey on
Bengaluru, India, Mar. 2019, pp. 1–6.
secure communication protocols for IoT systems,’’ in Proc. Int. Workshop
[19] A. Alkandari, I. F. Al-Shaikhli, A. Najaa, and M. Aljandal, ‘‘Accident Secure Internet Things (SIoT), 2016, pp. 47–62.
detection and action system using fuzzy logic theory,’’ in Proc. Int. Conf.
[45] M. Elkhodr, S. Shahrestani, and H. Cheung, ‘‘Emerging wireless technolo-
Fuzzy Theory Its Appl. (iFUZZY), Dec. 2013, pp. 385–390.
gies in the Internet of Things : A comparative study,’’ Int. J. Wireless Mobile
[20] Global Status Report on Road Safety 2018. Geneva: World Health
Netw., vol. 8, no. 5, pp. 67–82, Oct. 2016.
Organization; 2018. Licence: CC BYNC-SA 3.0 IGO. Accessed:
[46] O. Khutsoane, B. Isong, and A. M. Abu-Mahfouz, ‘‘IoT devices and
May 15, 2020. [Online]. Available: https://www.who.int/publications-
applications based on LoRa/LoRaWAN,’’ in Proc. 43rd Annu. Conf. IEEE
detail/global-status-report-on-road-safety-2018
Ind. Electron. Soc. (IECON), Oct. 2017, pp. 6107–6112.
[21] K. K. Patel, S. M. Patel, P. G. Scholar, and C. Salazar, ‘‘Internet of Things-
IOT: Definition, characteristics, architecture, enabling technologies, appli- [47] S. Al-Sarawi, M. Anbar, K. Alieyan, and M. Alzubaidi, ‘‘Internet of
cation & future challenges,’’ Int. J. Eng. Sci. Comput., vol. 6, no. 5, Things (IoT) communication protocols: Review,’’ in Proc. 8th Int. Conf.
pp. 6122–6131, 2016. Inf. Technol. (ICIT), May 2017, pp. 685–690.
[22] T. Mufti, N. Sami, and S. S. Sohail, ‘‘A review paper on Internet of Things [48] D. M. Ibrahim, ‘‘Internet of Things technology based on LoRaWAN rev-
(IoT),’’ Indian J. Appl. Res., vol. 9, no. 8, pp. 24–25, 2019. olution,’’ in Proc. 10th Int. Conf. Inf. Commun. Syst. (ICICS), Jun. 2019,
pp. 234–237.
[23] P. Asghari, A. M. Rahmani, and H. H. S. Javadi, ‘‘Internet of Things
applications: A systematic review,’’ Comput. Netw., vol. 148, pp. 241–261, [49] R. S. Sinha, Y. Wei, and S.-H. Hwang, ‘‘A survey on LPWA technology:
Jan. 2019. LoRa and NB-IoT,’’ ICT Express, vol. 3, no. 1, pp. 14–21, Mar. 2017.
[24] S. Lakshmipriya, ‘‘Exploring the nuances of Internet of Things in [50] M. Luvisotto, F. Tramarin, L. Vangelista, and S. Vitturi, ‘‘On the use
health care assisting system,’’ Int. Res. J. Eng. Technol., vol. 6, no. 2, of LoRaWAN for indoor industrial IoT applications,’’ Wireless Commun.
pp. 1776–1780, 2019. Mobile Comput., vol. 2018, May 2018, Art. no. 3982646.
[25] H. F. Atlam, R. J. Walters, and G. B. Wills, ‘‘Internet of Things: State-of- [51] W. Ayoub, A. E. Samhat, F. Nouvel, M. Mroue, and J.-C. Prevotet,
the-art, challenges, applications, and open issues,’’ Int. J. Intell. Comput. ‘‘Internet of mobile things: Overview of LoRaWAN, DASH7, and NB-IoT
Res., vol. 9, no. 3, pp. 928–938, Sep. 2018. in LPWANs standards and supported mobility,’’ IEEE Commun. Surveys
[26] R. Vignesh and A. Samydurai, ‘‘Security on Internet of Things (IoT) with Tuts., vol. 21, no. 2, pp. 1561–1581, 2nd Quart., 2019.
challenges and countermeasures,’’ Int. J. Eng. Develop. Res., vol. 5, no. 1, [52] A. Adhikary, X. Lin, and Y.-P.-E. Wang, ‘‘Performance evaluation of
pp. 417–423, 2017. NB-IoT coverage,’’ in Proc. IEEE 84th Veh. Technol. Conf. (VTC-Fall),
[27] R. Khan, S. U. Khan, R. Zaheer, and S. Khan, ‘‘Future Internet: The Inter- Sep. 2016, pp. 1–5.
net of Things architecture, possible applications and key challenges,’’ in [53] S. Madakam, R. Ramaswamy, and S. Tripathi, ‘‘Internet of Things (IoT):
Proc. 10th Int. Conf. Frontiers Inf. Technol., Dec. 2012, pp. 257–260. A literature review,’’ J. Comput. Commun., vol. 3, pp. 164–173, Jan. 2015.
[28] V. Sharma and R. Tiwari, ‘‘A review paper on ‘IOT’ & it’s smart applica- [54] B. Shabir, M. A. Khan, A. U. Rahman, A. W. Malik, and A. Wahid,
tions,’’ Int. J. Sci., Eng. Technol. Res., vol. 5, no. 2, pp. 472–476, 2016. ‘‘Congestion avoidance in vehicular networks: A contemporary survey,’’
[29] M. A. Khan, M. A. Khan, A. U. Rahman, A. W. Malik, and S. A. Khan, IEEE Access, vol. 7, pp. 173196–173215, 2019.
‘‘Exploiting cooperative sensing for accurate target tracking in industrial [55] W. Farooq, M. A. Khan, S. Rehman, and N. A. Saqib, ‘‘A survey of
Internet of Things,’’ Int. J. Distrib. Sensor Netw., vol. 15, no. 12, Dec. 2019, multicast routing protocols for vehicular ad hoc networks,’’ Int. J. Distrib.
Art. no. 155014771989220. Sensor Netw., vol. 11, no. 8, 2015, Art. no. 923086.
[30] V. N. Malavade and P. K. Akulwar, ‘‘Role of IoT in agriculture,’’ IOSR [56] D. B. Tushara and P. A. H. Vardhini, ‘‘Wireless vehicle alert and collision
J. Comput. Eng., vol. 1, no. 13, pp. 56–57, 2016. prevention system design using atmel microcontroller,’’ in Proc. Int. Conf.
[31] V. Subrahmanyam and K. Aruna, ‘‘Future automobile an introduction of Electr., Electron., Optim. Techn. (ICEEOT), Chennai, India, Mar. 2016,
IoT,’’ Int. J. Trend Res. Develop., vol. 2, pp. 88–90, Apr. 2017. pp. 2784–2787.
[32] S. Kawthankar and C. Raut, ‘‘A survey on smart automobiles using Internet [57] U. Patil, P. More, R. Pandey, and U. Patkar, ‘‘Tracking and recovery of
of Things for digital India,’’ Int. J. Recent Trends Eng. Res., vol. 3, no. 5, the vehicle using GPS and GSM,’’ Int. Res. J. Eng. Technol., vol. 4, no. 3,
pp. 27–33, 2017. pp. 2074–2077, 2017.
[33] R. K. Bajaj, M. Rao, and H. Agrawal, ‘‘Internet of Things (IoT) in the smart [58] B. S. Anil, K. A. Vilas, and S. R. Jagtap, ‘‘Intelligent system for vehicular
automotive sector: A review,’’ in Proc. IOSR J. Comput. Eng. (IOSR-JCE), accident detection and notification,’’ in Proc. Int. Conf. Commun. Signal
Conf. Recent Trends Comput. Eng. (CRTCE), 2018, pp. 36–44. Process., Melmaruvathur, India, Apr. 2014, pp. 1238–1240.
[34] M. U. Farooq, M. Waseem, S. Mazhar, A. Khairi, and T. Kamal, ‘‘A review [59] R. K. Megalingam, R. Nammily Nair, and S. Manoj Prakhya, ‘‘Wireless
on Internet of Things (IoT),’’ Int. J. Comput. Appl., vol. 113, no. 1, pp. 1–7, vehicular accident detection and reporting system,’’ in Proc. Int. Conf.
2015. Mech. Electr. Technol., Singapore, Sep. 2010, pp. 636–640.

122496 VOLUME 8, 2020

U. Alvi et al.: Comprehensive Study on IoT Based Accident Detection Systems for Smart Vehicles

[60] H. N. Kumar and G. Deepak, ‘‘Accident detection and intelligent naviga- MUAZZAM A. KHAN KHATTAK (Senior Mem-
tion system for emergency vehicles in urban areas using IoT,’’ Int. J. Eng. ber, IEEE) received the Ph.D. degree in com-
Tech., vol. 3, no. 6, pp. 330–334, 2017. puter sciences as a Sandwich Program from IIUI
[61] E. Nasr, E. Kfoury, and D. Khoury, ‘‘An IoT approach to vehicle accident and UMKC, Kansas City, USA, in 2011, and the
detection, reporting, and navigation,’’ in Proc. IEEE Int. Multidisciplinary Ph.D. degree from the University of Missouri,
Conf. Eng. Technol. (IMCET), Beirut, Lebanon, Nov. 2016, pp. 231–236. Kansas City, in 2016. He is currently working as
[62] Z. Wen and J. Meng, ‘‘Design of vehicle positioning system based on a Tenured Associate Professor with the Depart-
ARM,’’ in Proc. Int. Conf. Bus. Manage. Electron. Inf., Guangzhou, China,
ment of Computer Science, Quaid-i-Azam Uni-
May 2011, pp. 395–397.
versity, Islamabad, Pakistan. He joined the School
[63] M. Gaber, A. M. Wahaballa, A. M. Othman, and A. Diab, ‘‘Traffic acci-
dents prediction model using fuzzy logic: Aswan desert road case study,’’ of Electrical Engineering and Computer Science
J. Eng. Sci. Assiut Univ, vol. 45, no. 1, pp. 28–44, 2017. (SEECS), NUST, Islamabad, as an Assistant Professor, in 2013, where he
[64] X. Li, D. Lord, Y. Zhang, and Y. Xie, ‘‘Predicting motor vehicle crashes was promoted to the Associate Dean and a Tenured Associate Professor,
using support vector machine models,’’ Accident Anal. Prevention, vol. 40, in 2017. He has been with the School of Computer Science, University of
no. 4, pp. 1611–1618, Jul. 2008. Ulm, Germany, and the School of Computer and the Electrical Engineering,
[65] I. Ohe, H. Kawashima, M. Kojima, and Y. Kaneko, ‘‘A method for auto- University of Missouri, as a Research Fellow. He has worked at the Network-
matic detection of traffic incidents using neural networks,’’ in Proc. Pacific ing and Multimedia Laboratory, University of Missouri Kansas City, as a
Rim TransTech Conf. Vehicle Navigat. Inf. Syst. Conf. 6th Int. VNIS. Ride Research Fellow. His research interests include the Internet of Things, next-
Future, 1995, pp. 231–235. generation intelligent networks, block chain-based information and network
[66] B. Pan and H. Wu, ‘‘Urban traffic incident detection with mobile sensors security, vehicular ad-hoc/sensor actor networks, and acoustic/multimedia
based on SVM,’’ in Proc. 32nd Gen. Assem. Sci. Symp. Int. Union Radio networks.
Sci. (URSI GASS), Aug. 2017, pp. 1–4.
[67] Z. Halim, R. Kalsoom, S. Bashir, and G. Abbas, ‘‘Artificial intelligence
techniques for driving safety and vehicle crash prediction,’’ Artif. Intell. BALAWAL SHABIR received the bachelor’s
Rev., vol. 46, no. 3, pp. 351–387, Oct. 2016. degree in electrical engineering and the master’s
[68] S. Eduku, M. O. Alhassan, and J. Sekyi-Ansah, ‘‘Design of vehicle accident degree in information security from Air University
prevention system using wireless technology,’’ Int. J. Sci. Res. Publica- Islamabad. He is currently pursuing the Ph.D.
tions, vol. 7, no. 10, pp. 397–403, 2016. degree with NUST, Pakistan. His research inter-
[69] S. Kasera, ‘‘IoT based accident prevention system,’’ Int. J. Sci. Res. ests include cryptography and network security,
Develop., vol. 7, no. 5, pp. 832–835, 2019. privacy, machine learning, vehicular ad-hoc net-
[70] O. G. Basubeit, D. N. T. How, Y. C. Hou, and K. S. M. Sahari, ‘‘Distracted works, the Internet of Things, big data analyt-
driver detection with deep convolutional neural network,’’ Int. J. Recent
ics, blockchain, bioinformatics, and cloud and fog
Technol. Eng., vol. 8, no. 4, pp. 6159–6163, 2019.
computing.
[71] K. Dwivedi, K. Biswaranjan, and A. Sethi, ‘‘Drowsy driver detection
using representation learning,’’ in Proc. IEEE Int. Advance Comput. Conf.
(IACC), Feb. 2014, pp. 995–999. ASAD WAQAR MALIK (Senior Member, IEEE)
[72] M. Szarvas, A. Yoshizawa, M. Yamamoto, and J. Ogata, ‘‘Pedestrian detec- received the Ph.D. degree with majors in parallel
tion with convolutional neural networks,’’ in Proc. IEEE Intell. Vehicles and distributed simulation/systems from NUST,
Symp., Las Vegas, NV, USA, Jun. 2005, pp. 224–229. Pakistan, in 2012. He is currently an Assistant Pro-
[73] M. Murshed and M. S. Chowdhury, ‘‘An IoT based car accident prevention
fessor with the SEECS, NUST. He is also working
and detection system with smart brake control,’’ in Proc. Int. Conf. Appl.
as a Senior Lecturer with the Department of Infor-
Techn. Inf. Sci. (iCATIS), 2019, p. 23.
mation Systems, Faculty of Computer Science and
Information Technology, University of Malaya,
Malaysia. His primary research interests include
distributed simulation, cloud/fog computing, and
the Internet of Things.

SHER RAMZAN MUHAMMAD received the

Ph.D. degree from TU Berlin, Germany, in 2007.
In 1998, he joined International Islamic Univer-
UNAIZA ALVI received the bachelor’s degree in sity, Islamabad (IIUI), as an Assistant Professor.
information technology from the University of the From 2008 to 2012, he has served as a Profes-
Punjab, Lahore, in 2019. She is currently pursuing sor, the Chairman, and the Dean FBAS, from
the master’s degree in information technology with 2017 to 2018. He is currently a Professor with the
NUST, Islamabad. Her research interests include Faculty of Computing and Information Technol-
machine learning, the Internet of Things, vehicular ogy (FCIT), King Abdulaziz University Jeddah,
ad-hoc networks, and artificial intelligence. Saudi Arabia. He has authored more than 100 pub-
lications. His research interests include information security, IP multimedia
subsystem, WSN, healthcare, the IoT, and NGN.

VOLUME 8, 2020 122497