DAP391m

Network for Intelligent Parking System

SE183644 Nguyễ n Hàn Như
SE192567 Nguyễ n Duy Sơn
SE184292 Nguyễ n Hải Đăng
SE192307 Nguyễ n Lê Anh Quân
INTRODUCTION
Vietnamese License Plate Recognition (LPR) is
crucial for modern traffic management .
However, diverse plate designs and challenging
conditions in Vietnam present obstacles.

Using:
• WPOT- Net
• Tesseract OCR
 Define the problem
Objective: Accurate and efficient license plate recognition in Vietnam.
Key Tasks: Detecting license plates in images/videos.
Extracting text from detected plates.
Importance: Essential for modern parking systems.
Automates parking management, toll collection, and traffic
monitoring.
Benefits: Streamlines vehicle entry/exit in malls, hospitals, and
residential areas.
Reduces manual labor and improves operational efficiency.
 CHALLENGE
Diverse License Plates: Different colors for private,
government, and commercial vehicles require a robust
detection model.
Varying Lighting Conditions: Low-light at night and glare
during the day affect recognition accuracy.
Obstructions: Dirt, dust, and partial blockages complicate
plate detection.
Environmental Factors: Weather conditions like rain can
reduce visibility.
Technical Requirements: Advanced deep learning
techniques and high-quality dataset preparation are
needed for accurate recognition.
 Survey of Previous Solutions

Overview of Traditional Challenges in Vietnamese

Modern Advances in LPR
Approaches: Context

Traditional methods for Recent studies have Vietnamese license plates

License Plate Recognition integrated deep learning present unique challenges,
(LPR) relied on image models (e.g., YOLOv4, such as diverse designs
processing and machine (e.g., white, blue, yellow
CNN) with OCR (e.g.,
learning techniques (e.g., plates), varying lighting
HOG features with SVM
Tesseract) to improve conditions, and real-time
classifiers). These methods accuracy and robustness. processing requirements.
struggled with challenges
like low lighting, steep
angles, and occlusions.
Problem Analysis

Key Challenges:
High variance in license plate designs (e.g., private, government, and commercial vehicles).
Complex lighting conditions (e.g., low-light environments, glare, and shadows).
Real-time processing requirements for deployment in intelligent parking systems.
Existing Solutions and Their Limitations:
Previous solutions (e.g., YOLOv4, Tesseract OCR) have partially addressed these challenges but
still struggle with low-quality images, distorted characters, and heavy computational
requirements.
Selected Problem for Resolution:
The team has chosen to focus on improving the accuracy and robustness of LPR systems in
challenging Vietnamese conditions, particularly in low-light environments and with diverse
plate designs.
Proposed
Solution Advantages of the Proposed Solution:
High accuracy in detecting and recognizing
Solution Overview: license plates under challenging
The proposed solution combines WPod-Net for conditions.
license plate detection and Tesseract OCR for Real-time performance suitable for
character recognition. This hybrid approach deployment in intelligent parking systems.
addresses the challenges of Vietnamese LPR by Adaptability to diverse Vietnamese license
leveraging the strengths of deep learning and plate designs and environmental
traditional OCR. variations.
Solution Components: Disadvantages of the Proposed Solution:
a. WPod-Net for License Plate Detection: Computationally intensive, especially for
Detects license plates in various conditions high-resolution images.
(e.g., low light, occlusions). Performance depends on preprocessing
Outputs bounding boxes around detected steps (e.g., image enhancement).
plates. Requires a large, diverse dataset for
training and fine-tuning.
b. Tesseract OCR for Character Recognition:
Converts character images into text.
Supports Vietnamese language and diverse
fonts.
 Gaps
Lorem ipsum dolor sit amet, consectetur adipiscing elit,
sed do eiusmod tempor incididunt ut labore et dolore
magna aliqua. Ut enim ad minim veniam, quis nostrud
exercitation ullamco laboris nisi ut aliquip ex ea commodo
consequat.

Duis aute irure dolor in reprehenderit in voluptate velit

esse cillum dolore eu fugiat nulla pariatur. Excepteur sint
occaecat cupidatat non proident, sunt in culpa qui officia
deserunt mollit anim id est laborum.
Proposed solution
(overview)
Objective: Develop a high-accuracy, real-time LPR system tailored for
Vietnam.
Challenges Addressed: Diverse plate formats, environmental conditions,
and real-world obstructions.
Three Main Components:
License Plate Detection (WPOT-Net)
Text Recognition (Tesseract OCR )
Database Management
License Plate Detection
Uses WPOT-Net for real-time plate detection.
Quickly identifies and locates plates in various
conditions.
Filtering process applied to improve detection
accuracy.
 Text Recognition
Uses Tesseract OCR for character recognition.
Deep learning model with multiple layers for
better accuracy.
Handles different lighting, obstructions, and
viewing angles.
 Database Management & Performance
Optimization
Database: Stores and manages
vehicle information.
Data Augmentation: Enhances
dataset with brightness
adjustments, noise, and simulated
obstructions.
Performance Optimization:
Ensures real-time speed and high
accuracy.
Continuous Testing: Evaluates
system under different conditions.
 Result
Model Efficiency and Deployment
Model Size:
19MB model size, optimized for edge
Comparison:
devices.
YOLOv5+CRNN: 89.7% accuracy, 32
This is the model of automated access
FPS.
control gate for a smart parking lot
Our System: 84% accuracy

IR sensors detect vehicles or

pedestrians passing through.
A servo motor controls the barrier.
An LCD screen displays the system
status.
The ESP8266 WiFi module enables
remote management via an
application.
 Error Analysis
Lighting Variations Occlusions and Dirt Distorted
Motion Blur and
Characters:
Image Quality

Extreme lighting Partial obstructions, dirt, Poor image quality due to

conditions, such as or dust on license plates fast-moving vehicles or Incorrect recognition
monsoon rains and can reduce recognition focus issues led to occurred on plates with
tropical glare, still pose accuracy, particularly recognition errors. distorted characters due
challenges, especially in when the deformable to camera angles or
low-light environments WPod-Net backbone reflections.
where the Adaptive struggles to localize the The WPOD-NET system
Spatial Attention Module plate accurately. Partial obstructions, dirt, or dust on license plates faced challenges with
can reduce recognition accuracy, particularly when
may not fully the deformable YOLOv8 backbone struggles to oblique views, limiting the
localize the plate accurately.
compensate. detection of characters
and digits on transformed
plates
 Achievements
Conclusion The Hybrid WPod-Net and Tesseract OCR system
achieved 87% detection accuracy and 84%
character recognition accuracy on a custom
The paper concluded that combining WPod-Net and Vietnamese dataset of 3,059 images. It enhances
Tesseract OCR improves accuracy and efficiency in low-light performance with an Adaptive Spatial
Vietnamese license plate recognition but faces Attention Module and improves detection
challenges with low-resolution and distorted images. accuracy using WPod-Net.

Future Work
Future research will enhance OCR
accuracy, optimize real-time
processing, and integrate AI to
address issues with low-resolution
and distorted images.
Optimize for low-light conditions.
Expand dataset to include more
regional plate variations