International Journal of Science, Engineering and Technology Research (IJSETR)

Volume 6, Issue 4, April 2017, ISSN: 2278 -7798

REVIEW ON VEHICLE DETECTION

TECHNIQUES
Gopal Manne , Neetesh Raghuwanshi

 and dilation, and the quantiy of vehicles is large enough that

Abstract— Traffic has increased enormously with the economic area is to be selected and vehicles are identified in the form of
development. Due to the increasing urban population and hence bounding box from the image. So,blob analysis is used for the
the number of cars, need of controlling the traffic in streets, detection of number of vehicles each separately.
highways and roads is vital. Vehicle detection has been a part of In 2014, Prutha Y M et al. [4] proposed a system that
the traffic surveillance system for many years. Still the vehicle
detects the vehicle in real time in highway is done by using
detection method is challenging .In this paper, a system that
detects the vehicle in real time in highway is done by using image image processing. Each image is processed separately and the
processing. The implementation includes algorithms used for real number of cars has been counted by applying a threshold
time vehicle detection. The Vehicle detection is used to identify the value. If the number of cars increases beyond the value of
vehicles in any video or image file. The process of detection of threshold, then it automatically gives heavy traffic warning.
vehicles includes the object detection by considering the vehicles In 2015, PAWAR B.D et al. [5] introduces a comparison in
as the primary object. Now a days vehicle detection techniques are the study of automatic Vehicle detection from high resolution
developing and existing techniques are improving to get better aerial images. The approach is based on morphological
accuracy results. This methods mostly used in traffic monitoring, operation that describes the detection using structure based
speed management and also in military applications. The objective
representation. During feature extraction, the study compares
of this paper is to present a survey of research related to vehicle
detection techniques and the application of vehicle detection the two basic morphological operations like Top-Hat and
techniques for traffic management and other applications such as Bottom-Hat. The experimental result shows promising result
in highway traffic surveillance control, management and urban using proposed method and gives precise detection of
traffic planning. vehicles. This paper identifies a proficient and robust method
for automatic vehicle detection from highway aerial images
Index Terms— vehicle detection, Feature extraction. among two significant morphological operations (Top-Hat &
Bottom-Hat). We use morphological methods as image
1) RELATED WORK pre-processing, Top/Bottom Hat processing, Identification of
In 2013, R.Sindoori et al. [1] suggested an approach for Vehicle shape, Segmentation of vehicles to identify
vehicle detection system from satellite images. Vehicle candidate vehicles. From This ,we can say that the Top-Hat
detection is done by pixel wise classification method. The operation is the efficient technique for vehicle detection.
essential part of the paper is to extract the feature and classify In 2010, Yan Liu et al.[6] proposed, a Quasi-shot
the vehicle color. Extraction of feature involves edge and Segmentation Vehicle Detection (QSVD) method to improve
corner detection. For edge detection, the Canny edge detector the accuracy and robustness of traffic flow detection. The
technique is applied.The Harris corner detector process is QSVD method is based on histogram intersection method to
applied for the corner detection.Adaboost is used for find the biggest difference between successive frames.
extraction of vehicle colour to divide vehicle and non-vehicle Adjacent frames are considered in when vehicle is detected in
colours. An absolutely, morphological operations are applied current frame. Relying on the QSVD method, vehicle can be
for the enhancement of the vehicle detection. detected by signal peak and valley search and gradual
In 2013, Youpan Hu et al. [2] proposed the Haar features transition shot detection problem can be resolved. The initial
and Histograms of Oriented Gradients (HOG) as helpful video is used to accomplish detection directly without image
features for the detection of vehicle respectively. They preprocessing.
propose a method proposed a method to to detect vehicles in In 2014, Bharti Sharma et al. [7] proposed differential
videos and classify them into two types depending on morphology closing profile to extract the vehicle
combined the Haar features and HOG features. Due to this automatically from the traffic image. They also made
method,it can classify and detect the vehicles in comparison from previous image processing based methods
multi-orientations with good classification results. and the results indicate that the proposed method provides
In 2015, Ekta Saxena et al. [3] proposed a algorithm for better results than traditional image processing based
Object detection from a satellite image or aerial image and methods.
developed automated system using morphological
recognition algorithm in MATLAB R2013a in which image In 2015, RON MAHABIR et al. [8] proposed a technique
is captured from camera and converted into gray scale image for the detection and identification of vehicles from RGB
for pre-processing.The binary conversion is applied on image images, by making use of both object spectra and image
after completion of image complemented through morphology. The identification performance results increases
conversion.After conversion canny edge detection method upto 62% with false positives occurring by using the images
then it is passed through the dilation process. After filtration with sun glare and vehicles with like spectra values.

482
All Rights Reserved © 2017 IJSETR
 International Journal of Science, Engineering and Technology Research (IJSETR)
Volume 6, Issue 4, April 2017, ISSN: 2278 -7798

In 2012, Zezhong Zheng et al.[9] proposed a novel and

robust method for automatic vehicle detection from highway
aerial image. In this method, a GIS road vector map is used to
limit a detection of vehicle system to the highway
management networks. Here, morphological oerationaal
method is used to identify candidate vehicles. Experiment is
conducted with 0.15 m resolution aerial image. And the result
demonstrated that the novel method has an excellent
detection performance, thus the method is very promising.
In 2016, Sundaresh Ram et al. [10] proposed an automated
method for detecting vehicles of varying sizes in
low-resolution aerial imagery. First, they develop a new
vehicle enhancement filter involving multiscale Hessian
analysis. After thresholding, they refine the candidate vehicle
detections based on analysis of bilateral symmetry. They
shows that the proposed method provides improved detection
accuracy compared with existing vehicle detection
algorithms for various low-resolution aerial images.
In 2010, Isha Jain et al.[11] developed an algorithmic
approach to vehicle detection and classification using fuzzy
logic. This not only reduces the complexity of the system but
enhances its use in the areas which are too difficult to be
detected by normal means. Further, it is proposed that after
detection objects can be classified using techniques like
neuro-fuzzy etc. so as supervised and unsupervised learning
can be used to train the system. This algorithm can be applied
on real time projects. They had taken the images of moving
and still vehicles and an algorithm is used for vehicle
detection.
In 2015, A. Shakin Banu et al. [12] proposed
morphological operations and Histogram of Gradient (HOG)
feature extraction for detection of vehicle.By finding the
common pixels in the detected edge and the ROI regions,we
The frame width and height are 320 x 240. The data rate for
can obtain the gradient.The algorithm (HOG) is an efficient
video is 2800kbps. Fig.2(a) represents original image and
feature extraction for detecting the vehicles whose result is
2(b) shows background removal. The histogram
extremely helpful for traffic analysis, management and
representation of the image is shown in the fig. 2(c). Edge is
surveillance. By identifying the vehicle size,the future work
detected after removing background colour &representation
can be improved and the size of identified object is belongs to
of histogram. Canny edge detector detects the edge of image
truck or car could be found by comparing indentified size
is shown in the fig. 3(b). The purpose of Harris corner
with threshold value. The a success rate of around 83% of
detector to detects the corner.The vehicle detection using
accuracy is achieved in vehicle detection using proposed
adaboost is shoen in fig. 3(c )
method.
Authors in [2] have been considered Moving object
detection and classification which is a key task in video
monitoring applications. Authors uses real videos to test the
2) PROBLEM FORMULATION result of proposed method and compute the accuracy rate.
Authors in [1] have been considered experimental results
as shown in Fig.1 which shows the snapshots of experimental
images. The experimental results show better performance
and flexibility.

They used Haar a feature vector to represent a vehicle, and

the results show that Haar features can describe images well
and fast. Using AdaBoost, weak classifiers can be combined
to create a strong classifier. In this method, HOG features can
describe as different types of vehicles, and a strong classifier
can distinguish them from each other.In this experiment,
author chooses 2530 positive images contain vehicle and

483
All Rights Reserved © 2017 IJSETR
 International Journal of Science, Engineering and Technology Research (IJSETR)
Volume 6, Issue 4, April 2017, ISSN: 2278 -7798

4960 negative images contain random backgrounds(Fig.4), Authors in [4] have been considered the algorithm is tested
which got from all kinds traffic videos. under various traffic conditions, using MATLAB & various
methods are applied which gives the output and is shown in
the figure Fig 6 (a) and Fig 6( b). In fig. 6(a) the left most
figure shows the image of the highway when initially there is
no traffic and the middle image shows the background
subtracted image where no object is present and the rightmost
image shows the edge detector applied and the bottom figure
shows 0km/h which is the speed of the cars as there is no
traffic movement so the arrival rate is also 0 here.

As is shown in Table-1, accuracy rate reaches

97.2%, and most of the vehicle can be detected except the
vehicles which size is too small or too large. When this
algorithm is applied in vehicle detection, about 95.4% of
vehicles can be detected in one frame image. After the step of
vehicle detection, they got the images of vehicles, and there is
only one vehicle in an image. All of the images are resize into
the size of the same 64*64 pixels, which are shown in fig.4
.They built our vehicle image database with 2550 “small
vehicles” and 1440 “oversize vehicles” respectively for
training. All the images are detected from real world highway
monitor video. And the last step, they used some testing
images to count the testing accuracy rate. The results of the
experiment are listed in the Table-2. The HOG features work
well in vehicle type classification, the classification accuracy
is quite high with a large amount of test samples. And as all
the vehicles are detected from the traffic surveillance video
Fig.5 shows oversize vehicle samples and small vehicles
samples

the figure Fig 6 (b) the leftmost figure shows the image of the
highway when there is traffic and cars travelling on the road
and the middle image shows the background subtracted
image where objects are detected hence it is in white patch
and the background is made black and the rightmost image is
the edge detector applied which has different edge detection
techniques applied to it and optimal detector is selected and
the figure in the bottom shows the objects being tracked and
masked with rectangular boxes and numbered as the object
arrives which is considered as frames. The speed and the
arrival rates of the cars are estimated by setting a threshold
value. As the speed n arrival rate changes the values are
shown on the screen. The bottom figure shows the
performance graph of the traffic flow where the x-axis is the
number of blocks and the y-axis the percentage of motion of
the vehicles, it varies as the traffic movement changes.
Authors in [5] have been considered total 9 numbers
are cars present shown in following fig. In the vehicle
detection, total vehicle detected are 9 using Top-Hat
processing and Bottom-Hat processing. All the testing result
is applied on 5 highways. Fig 7(A) to 7(G) shows result of
Top-Hat processing as which are as follows,

484
All Rights Reserved © 2017 IJSETR
 International Journal of Science, Engineering and Technology Research (IJSETR)
Volume 6, Issue 4, April 2017, ISSN: 2278 -7798

Authors in [9] have been stated that the bot-hat and

top-hat images can be partitioned with Ostu method (see fig
10), and fig11). And the vehicles can be detected
respectively. The detected vehicles with two different
methods could be overlapped (see fig 12). As the methods
described, from fig.12, total 17 cars was detected, only one
car was omitted and one false car objection was detected.
Therefore, the precision is very high.

Authors in [7] have been considered The automated

vehicle detection using morphological differential profile is tested
on image . The detection algorithm is tested on the traffic image as
shown in fig 8. Total vehicles visible in the input image are 26. The
algorithms [15] [16] is tested on the same traffic image and the 3) CONCLUSION
statistics measures of proposed and other algorithms [15] [16] are
This paper provides a summarizing study on the
summarized in Table 1, which present the false positive rate, true
positive rate, false negative rate .The statistics measures of Table-3 proposed techniques which have used in traffic video. It
indicate that the proposed vehicle detection method has better focuses in on vehicle detection This review shows the
detection rate than other methods. Figures fig 8 & fig.9 show how these vehicle detection methods plays an important
original image, the result of method [14] [15] and proposed role in the traffic surveillance systems gives better
algorithm respectively. Method [14] result shows that some detected understanding and highlights the issues and its solutions
vehicles are very blurred and merged with background .So it is for traffic surveillance systems.
difficult to recognize them as true vehicles. The true positive rate of
the method [14] is 84%, method [15] is 38% and proposed algorithm REFERENCES
shows 96% true vehicle detection rate.
1) R.Sindoori, K.S .Ravichandran, B. Santhi, Adaboost Technique
For Vehicle Detection In Aerial Surveillance, International
Journal Of Engineering And Technology (Ijet), Vol.5 No. 2, Pp
765-769, Apr-May 2013
2) Youpan Hu, Qing He, Xiaobin Zhuang, Haibin Wang, Baopu Li,
Zhenfu Wen ,Bin Leng, Guan Guan, Dongjie Chen, Algorithm
For Vision-Based Vehicle Detection And Classification,
Proceeding Of The Ieee International Conference On Robotics
And Biomimetics (Robio) Shenzhen, China, Pp
568-572,December 2013.

485
All Rights Reserved © 2017 IJSETR
 International Journal of Science, Engineering and Technology Research (IJSETR)
Volume 6, Issue 4, April 2017, ISSN: 2278 -7798

3) Ekta Saxena, Mrs. Neha Goswami, Automatic Vehicle Detection

Techniques In Image Processing Using Satellite Imaginary,
International Journal On Recent And Innovation Trends In
Computing And Communication Volume: 3 Issue: 3,Pp 1178 –
1181, March 2015.
4) Prutha Y M, Anuradha S G, Morphological Image Processing
Approach Of Vehicle Detection For Real Time Traffic Analysis,
International Journal Of Engineering Research & Technology
(Ijert), Vol. 3 Issue 5, Pp 1452-1456,May – 2014.
5) Pawar B.D, Humbe V.T, Morphology Based Composite Method
For Vehicle Detection From High Resolution Aerial Imagery,
Vnsgu Journal Of Science And Technology, Vol.4,No.1,Pp
50-56, July-2015.
6) Yan Liu, Yaping Dai, Meng Wang, Zhongjian Dai, Zubair Ahmed
Memon, Real-Time Vehicle Detection Based On Quasi-Shot
Segmentation, 3rd International Congress On Image And Signal
Processing, Pp 447-450,2010
7) Bharti Sharma, Vinod Kumar Katiyar, Arvind Kumar Gupta,
Akansha Singh, The Automated Vehicle Detection Of Highway
Traffic Images By Differential Morphological Profile, Journal Of
Transportation Technologies, 4, 150-156,2014.
8) Ron Mahabir, Kevin Gonzales, Jared Silk, Vdis: A System For
Morphological Detection And Identification Of Vehicles In Rgb
Images, Journal Of Mason Graduate Research Vol. 2 No. 2, Pp
84-97,Year 2015.
9) Zezhong Zheng, Xiaoting Wang, Guoqing Zhou, Ling Jiang,
Vehicle Detection Based On Morphology From Highway Aerial
Images, Igarss(Ieee),Pp 5997-6000,2012.
10) Sundaresh Ram, Jeffrey J. Rodriguez, Vehicle Detection In Aerial
Images Using Multiscale Structure Enhancement And
Symmetry,Icip(Ieee),Pp 3817-3821,2016.
11) Isha Jain,Babita Rani , Vehicle Detection Using Image Processing
And Fuzzy Logic, International Journal Of Computer Science &
Communication, Vol. 1, No. 2, Pp. 255-257, July-December
2010.
12) A. Shakin Banu, P. Vasuki, Video Based Vehicle Detection Using
Morphological Operation And Hog Feature Extraction, Arpn
Journal Of Engineering And Applied Sciences, Vol. 10, No. 4,Pp
1866-1871, March 2015.
13) Http://Resources2.News.Com.Au/Images/2012/03/28/1226312/31
5410-Southern-Expressway-Duplication.Jpg
14) Brown, L.M., Senior, A.W., Tian, Y.L., Connell, J., Hampapur,
A., Shu, C.F., Merkl, H. And Lu, M. (2005) Performance
Evaluation Of Surveillance Systems Under Varying Conditions.
Ieee International Workshop On Performance Evaluation Of
Tracking And Surveillance, Colorado, January.
15) Zheng, Z.Z., Wang, X.T., Zhou, G.Q. And Jiang, L. (2012)
Vehicle Detection Based On Morphology From Highway Aerial
Images. 2012 Ieee International Geoscience And Remote Sensing
Symposium (Igarss), Munich, 22-27 July 2012, 5997-6000.
16) Ekanayake, E.M.C.L., Wijayakulasooriya, J.V. And Alahakoon,
P.M.K. (2012) Controlling High-Traffic Intensities: Detection Of
Vehicular Movement In A Pre-Specified Region Of Interest
Using Morphological Image Processing. International Journal Of
Scientific And Research Publications, 2, Issn2250-3153

Gopal Manne is Student at Department of Electronics and

Communication Engineering from RKDF Institute of Science and
Technology, Bhopal(MP), India. He is pursuing his M. Tech. in Digital
Communication. His area of interest is Image Processing.

Neetesh Raghuwanshi is Assistant Professor at Department of

Electronics and Communication Engineering from RKDF Institute of
Science and Technology, Bhopal(MP), India. He has completed M. Tech. in
Digital Communication. His area of interest is Image Processing.

486
All Rights Reserved © 2017 IJSETR