Application of image processing for the development

of Automatic Inspection system

National iConference iOn iAdvances iin iElectronics iSignal iProcessing iand iCommunications i(AESPC-2020)
 1
Aman Prasad Chennuri
dept. ECE 2
Sadanand Sagar Ramarapu 3
Mukesh Reddy Yanala
( btech) dept. ECE dept.ECE
Sreenidhi institute of science and (btech) (btech)
technology iSreenidhi institute of science and iSreenidhi institute of science and
(jntuh) technology technology
Hyderabad, India (jntuh) (jntuh)
[email protected] Hyderabad, India Hyderabad, India
[email protected] [email protected]

4
Lavanya E 5
DR. C. N. Sujatha
dept. ECE idept.ECE i
Assistant Professor Professor
Sreenidhi institute of science and iSreenidhi iinstitute iof iscience iand
itechnology
technology
(jntuh)
(jntuh) Hyderabad, iIndia
Hyderabad, India [email protected]
[email protected]
 Abstract—In manufacturing industry, machine vision is Machine vision image processing methods include:
very important nowadays. Computer vision has been
 Pixel counting: Counts the number of light or dark
developed widely in manufacturing for accurate automated
pixels.
inspection. A model of automated inspection system is
presented in this conceptual paper. Image processing is used  Thresholding: converts an image with gray tones to
for inspection of part. It is assumed that the part after going simpky black and white or using separation based on
through many previous operations comes to inspection system a grayscale value.
where the weight of the part as well as geometry made on that  Segmentation: Partitioning a digital image into
part is detected and later decided whether it is to be accepted multiple segments to simplify and/or change the
or rejected with the help of image processing technique. The representation of an image into something that is
image processing technique used for Defect detection and more meaningful and easier to analyze.
algorithms developed for defect detection and linear dimension
 Blob discovery & manipulation: Inspecting an image
measurement. Various types of sensors were interfaced with
for discrete bobs of connected pixels (e.g., a black
the vision hardware and the part handling mechanism, to hole in a grey object) as image landmarks. These
complete the total automated vision based inspection system. blobs frequently represent optical targets for
This system is an accurate, repeatable, fast and cheap solution machining, robotic capture, or manufacturing failure.
for industries. Using MATLAB software a program is
developed and pattern or geometry is detected.
 Pattern recognition including template matching:
Finding, matching, and/or counting specific patterns.
Keywords- Matlab, machine vision, computer vision, image This may include location of an object that may bw
processing. rotated, partially hidden by an another object, or
varying in size.
I. INTRODUCTION
II. EXISTING STANDARDS
Automated inspection systems are continuously
conveyed in the manufacturing process. The systems are A. Imagenet large scale visual recognition challenge
capable of measuring predetermined parameters of various The Imagenet large scale visual recognition challenge is
parts, comparing the measured parameters with a benchmark in object category classification and detection
predetermined values, evaluating from the measured of hundreds of object categories and millions of images. The
parameters the integrity of the parts and determining whether challenge has been run annually from 2010 to present,
such parts are acceptable or, alternatively, whether they attracting participation from more than fifty institutions.
should be rejected.Humans are able to find such defects This paper describes the creation of benchmark dataset and
with prior knowledge. Human judgment is influenced by the advances in object recognition that have been possible as
expectations and prior knowledge. However, it is tedious, a result. We discuss the challenges of collecting large-scale
laborious, costly and inherently unreliable due to its ground truth annotation, highlight key breakthroughs in
subjective nature. Therefore, traditional visual quality categorial object recognition, provide a detailed analysis of
inspection performed by human inspectors has the potential the current state of the field of large-scale image
to be replaced by computer vision systems. The increased classification and object detection, and compare the state-of-
demands for objectivity, consistency and efficiency have the-art computer vision accuracy with human accuracy. We
necessitated the introduction of accurate automated conclude with lessons learned in the 5 years of the
inspection systems. These systems employ image processing challenge, and ropose future directions and improvements.
techniques and can quantitatively characterize complex
sizes, shapes, and the color and textural properties of B. Joint face detection and alignment using multitask
products. Accurate automated inspection and classification cascaded convolutional networks
can reduce human workloads and labor costs while Face detection and alignment in unconstrained
increasing the throughput. Machine vision has been used to environment are challenging due to various poses,
detect the part and take the image of the part which illuminations, and occlusions. Recent studies show that deep
compares it with the standard dimensions given to it through learning approaches can achieve impressive performance on
programming language these two tasks. In this letter, we propose a deep cascaded
Machine Vision the process of applying a range of multitask framework that exploits the inherent correlation
technologies and methods to provide imaging-based between detection and alignment to boost up their
automatic inspection, process control and robot guidance in performance. In particular, our framework leverages a
industrial applications. the first step in the MV sequence of cascaded architecture with three stages of carefully designed
operation is acquisition of an image, typically using cameras, deep convolutional networks to predict face and landmark
lenses, and lighting that has been designed to provide the location in a coarse-to-fine manner. In addition, we propose
differentiation required by subsequent processing. MV a new online hard sample mining strategy that further
software packages then employ various digital image improves the performance in practice. Our method achieves
processing techniques to extract the required information, superior accuracy over the state-of-the-art techniques on the
and often make decisions (such as pass/fail) based on the challenging face detection dataset and benchmark and
extracted information. Though the vast majority of machine WIDER FACE benchmarks for face detection, and
vision applications are still solved using 2 dimensional annotated facial landmarks in the wild benchmark for face
imaging, machine vision applications utilizing 3D imaging alignment, while keeps real-time performance.
are growing niche within the industry.

National iConference iOn iAdvances iin iElectronics iSignal iProcessing iand iCommunications i(AESPC-2020)
C. Deep convolutional network cascaded for facial point vision applications, the computers are pre-programmed to
detection solve a particular task, but methods based on learning are
We propose a new approach for estimation of the now becoming increasingly common.
positions of facial key points with three-level carefully IV. IMPLEMENTATION
designed convolutional networks. At each level, the outputs
of multiple networks are fused for robust and accurate Matlab using Toolboxes like image processing,
estimation. Thanks to the deep structures of convolutional Computer vision and Machine vision.
networks, global high-level features are extracted over the
whole face region at the initialization stage, which help to
locate high accuracy key points. There are two folds of
advantage for this. First, the texture context information
over the entire face is utilized to locate each key point.
Second, since the networks are trained to predict all the key
points simultaneously, the geometric constraints among key
points are implicitly encoded. The method therefore can
avoid local minimum caused by ambiguity and data
corruption in difficult image samples due to occlusions,
large pose variations, and extreme lightings. The networks
at the following two levels are trained to locally refine initial
predictions and their inputs are limited to small regions
around the initial predictions. Several network structures
critical for accurate and robust facial point detection are
investigated. Extensive experiments show that our approach
outperforms state-of-the-art methods in both detection
accuracy and reliability.
D. Real-time expression transfer for facial reenactment
We present a method for the real-time transfer of facial
expressions from an actor in a source video to an actor in a
target video, thus enabling the ad-hoc control of the facial
expressions of the target actor. The novelty of our approach
lies in the transfer and photorealistic re-rendering of facial
deformations and detail into the target video in a way that
the newly-synthesized expressions are virtually
indistinguishable from a real video. To achieve this, we
accurately capture the facial performances of the source and
target subjects in real-time using a commodity RGB-D
sensor. For each frame, we jointly fit a parametric model for
identity, expression, and skin reflectance to the input color
and depth data, and also reconstruct the scene lighting. For
expression transfer, we compute the difference between the
source and target expressions in parameter space, and
modify the target parameters to match the source
expressions. A major challenge is the convincing re-
rendering of the synthesized target face into the
corresponding video stream. This requires a careful
V. RESULTS AND DISCUSSION
consideration of the lighting and shading design, which both
must correspond to the real-world environment. We We have inspected objects like bricks, clips, rubbers and for
demonstrate our method in a live setup, where we modify a their size, shape, color, missing parts and dimensions.
video conference feed such that the facial expressions of a
Instead of using complicated filters like correlation and edge
different person (e.g., translator) are matched in real-time.
enhancement we analyzed the objects following the simplest
III. PROPOSED WORK and fastest approach possible, that is, inspect their grey scale
In our proposed system, we used industrial machine and RGB pixel values. This approach not only saves time
vision systems which, say, inspect bottles speeding by on a but it also proves that machine vision is possible without
production line, to research into artificial intelligence and application of complicated filters.
computers or robots that can comprehend the world around
them. The computer vision and machine vision fields have VI. CONCLUSION
significant overlap. Computer vision covers the core
technology of automated image analysis which is used in In this paper, we propose a new anchor-based cascade
many fields. Machine vision usually refers to a process of framework for efficient face detection, i.e., Anchor Cascade,
combining automated image analysis with other methods by exploring the multi-scale anchors in CNN-based cascade
and technologies to provide automated inspection and robot
guidance in industrial applications. In many computer- face detection framework. To further improve the recall rate,
we devise a context pyramid maxout mechanism in

National iConference iOn iAdvances iin iElectronics iSignal iProcessing iand iCommunications i(AESPC-2020)
harmony with the anchor cascade framework. By using the [1] iRafael iC. iGonzalez, i“Digital iImage iProcessing. i
proposed anchor cascade face detector, we further bridge the [2] iJohn iC. iRuss, i“The iImage iProcessing iHandbook”.
[3] iKenneth iR. iCastleman, i“Digital iImage iProcessing”.
gap between anchor-based face detectors and CNN-based
[4] l iBovik, i“Handbook iof iImage nd iVideo iProcessing”.
cascade face detectors. Specifically, the proposed anchor [5] iK. iMikolajczk, . iZisserman nd iC. iSchmid“Shape irecognition iwith
cascade face detector is comparable with typical CNN-based iedge-based ifeatures”. iIn iBritish iMachine iVision iConference,
iSeptember i2003
cascade face detectors, e.g., MTCNN, in both the model size
[6] iR. iKrishnan, i“Electric iMotor iDrives: iModeling, nalysis, nd
and the running speed, while the detection accuracy has iControl”. i
been greatly improved, e.g., from 0:9435 to 0:9704 at 1k [7] iF. iRothganger, iS. iLazebnik, iC. iSchmid nd iJ. iPonce.
false positives In FDDB. Experimental results on two i“Segmenting, imodeling nd imatching ivideo iclips icontaining
imultiple imoving iobjects”.IEEE iConference ion iComputer iVision,
popular face detection benchmarks, FDDB and WIDER i2004. i
FACE, demonstrate the effectiveness of the proposed anchor [8] iDavid iBenson, i“EASY iSTEP'n, n iIntroduction ito iStepper iMotors
ifor ithe iExperimenter ifrom iSquare i1 iElectronics”
cascade framework for efficient face detection

REFERENCES

National iConference iOn iAdvances iin iElectronics iSignal iProcessing iand iCommunications i(AESPC-2020)
National iConference iOn iAdvances iin iElectronics iSignal iProcessing iand iCommunications i(AESPC-2020)