Traffic Enhancement System Using Machine

Learning
1
Rajshri Tayde, B.Tech - Information Technology, VIIT Pune
2
Bhavesh Patil, B.Tech - Information Technology, VIIT Pune
3
Avinash Magar, B.Tech - Information Technology, VIIT Pune
4
Kiran Suryavanshi, B.Tech - Information Technology, VIIT Pune
5
Suvarna Pawar, Professor, Information Technology Department, VIIT Pune

Abstract

In today's world, two-wheelers and four-wheelers are the most preferred mode of transport. Motorcycle & Car accidents have
been rapidly growing within these years in many countries. Due to several social and economic factors, this type of vehicle is
becoming increasingly popular. The helmet is the main safety tool for motorcyclists but many drivers don’t use them. We present
an automatic method that uses an image processing technique by which motorcyclists not wearing helmet can be detected. This
moving vehicle can be detected by thresholding and then classified into motorcyclists and non-motor cyclists by area and aspect
ratio. If in case a motorcyclist is detected without a helmet, the number plate of a motorcycle is read and noted. A simple
algorithm is designed that can help to recognize number plates of motorcyclists using images taken by a camera. There are 5 parts
in recognition of the number plate algorithm: image procurement, preliminary processing, fringe detection and segmentation,
feature extraction, and recognition of character number plates using suitable machine learning algorithms. A database will be
generated with the records to identify every person who breaks the rule. After identifying the offender automatically money
deduct from his bank account and he gets system generated slip registered mobile number and email.

Keywords: Computer Vision, Machine Learning, Tracking, road safety, deep learning

I. INTRODUCTION

Motorcycles are widely used as a mean of transportation in many countries. Due to their major advantages such as low price and
low operation cost as compared to other vehicles. In the last decade, increase in the no. of accidents is being observed. As per the
Indian ministry of transport, in 2016 nearly 28 two wheeler riders died daily because of not wearing helmet. In 2017, 31 out of
100 people died in road accidents which lead to the increase in rate from 21.6 deaths per 100 accidents as compared to 2005.
Every year there are 1.4 million Traumatic Brain Injuries (TBI’s) in India. Around 76.5 billion dollars is spent in treatment
related to these injuries. More than 50,000 individuals died from TBI. This leaded to proposal for aiming at the security and
safety of the motorcyclists against the road accidents.

The main safety equipment of motorcyclist is the helmet. It protects motorcyclist against accidents. Although using helmet while
driving a two-wheeler is compulsory in many countries, but there are some drivers who do not use it or use incorrectly. The
department of transportation in USA said that, in 2011, only 66% of motorcyclist used helmet according to the law.

Over the past years many works were carried out in traffic analysis in various domains, including vehicle detection and
classification, and helmet detection. Intelligent traffic systems were usually implemented using vision computer algorithms, such
as: background and foreground image detection to segment moving objects in scene and image descriptors to extract features.
Computational intelligence algorithms are used too, like machine learning algorithms to classifier the objects. Next, some related
works to helmet detection are presented.

The system proposes an occlusion detection and segmentation method. The method uses the visual length, visual width, and pixel
ratio to detect the classes of the motorcycle occlusions and segment the motorcycle from each occlusive class. Leelasantitham et
al. [1] proposed a technique that detects moving vehicles using tracking methods. Also, Zengqiang et al. [8] presented a system
that identifies vehicles even if part of it is occluded. The system can track the detected vehicle even if the occlusion continues to
occur. In order to do this, a vehicle segmentation algorithm, which is based on feature points on contour, is presented.

II. LITERATURE SURVEY

The current work that solves the problem by image processing solutions uses technologies like HOG, LBP, WT [2][3][5].
Chiverton [2] proposed the use of circular arc to identify helmet in a video feed and has very less accuracy. On the other hand
given the number of vehicles on the speed at a given instance, the computation that required is very heavy and consumes lots of
resources. This method will determine any circular object around the bike rider as helmet.

Wen et al. [4] suggested a circle arc detection method that is based on the circular Hough transformation. They applied it to
detect helmet on the surveillance system of the Automatic Teller Machine. The weakness of this work is that they only use
geometric features to verify if any safety helmet exists in the scene. Geometric features are not enough to find helmet. The head
can be mistaken with a helmet because both had a circular-like form.

In another paper Chiu et al. [8] proposed a computer vision system aims to detect and segment motorcycles partly occluded by
another vehicle. A helmet detection system is used, and the helmet presence determines that there is a motorcycle in a scene. In
order to detect the helmet presence, the image edges are computed in the possible helmet region. The Canny edge detector is
used. The quantities of edge points which are similar to a circle define a helmet region. The method needs so much information
(helmet radius, camera angle and height) that must be provided by user.

Amit Verma et al. [6] proposed approach where image was stored in the form of matrix and the output was displayed in the form
of detected numbers. The overall work was to use Sobel Fringe detection technique. The result showed that it was the simplest of
all and with lesser complexity to detect the number.

Zengjia Yan [8] presented a novel method for vehicle shadow segmentation in a sequence of traffic images taken from a
stationary camera mounted on a tall building or pedestrian crossings bridge. Knoeppel [9] proposed a robust vehicle detection
system that could detect vehicles in the rear view of the host car. Two cameras with 12" focal length mounted behind the back
pane are used. Reliable vehicle recognition up to 150m and the exact track localization of the vehicles are the key features of the
system.

Takahashi et al. [9] implemented a computer vision system for bicycle, pedestrian and motorcycle detection. The system detects
moving objects (horizontal motion) and pedaling movement (vertical motion) using the Gabor filtering. The HOG descriptor is
computed and objects are classified into two classes, two-wheel vehicle and pedestrian, using SVM. At the end, the vertical
motion is computed and the two-wheel vehicles are classified into motorbike and bicycle. Sonoda et al. [10] proposed a system to
detect moving objects. The system aims to detect moving objects at an intersection (like vehicles and pedestrians) to warn the
driver. They used the mixture of Gaussians to detect the moving objects and the Lucas-Kanade Traker algorithm for pedestrian
tracking.

Sorin draghici proposed an artificial neural network based artificial vision system is able to analyze the image of a motorcycle
given by camera, locate the registration plate and recognize the registration number of the motorcycle. The system proposed by
Sorin draghichi et al., has designed using a modular approach which allows easy upgrading and /or substitutions of various sub
modules thus making it potentially suitable in large range of vision applications.

III. PROPOSED ARCHITECTURE

To detect whether a motorcyclist is wearing a helmet or not, we must identify a moving motorcycle. To do that, we first need to
classify the vehicles into 2 groups, motorcycles and other vehicles. Then, for each motorcycle, we consider the top part of the
area. If the motorcyclist is wearing a helmet, the top-part should be nicely rounded.
3.1 Threshold Method
Threshold is one of the widely methods used for image segmentation. It is useful in discriminating foreground from the
background. By selecting an adequate threshold value T, the gray level image can be converted to binary image. The binary
image should contain all of the essential information about the position and shape of the objects of interest (foreground). In this
paper, we choose a mutliple threshold technique as follows.

3.2 Mathematical morphology method

In this step, a noise is removed using binary mathematical morphology that is based on two basic operations, defined in terms of
a structuring element, a small window that scans the image and alters the pixels in function of its window content: a dilation of
set A with structuring element B enlarges the objects (more 1-pixels will be present in the image), an erosion shrinks objects (the
number of 1-pixels in the image decreases).

3. Feature Extraction
In this step, features of a detected moving area are extracted so that they can be used to decide whether moving area is a
motorcycle or not. In this paper, we use the area of the region and its rectangular shape as the features.

Fig 1. Motorcyclists with and without helmet

Architecture Diagram :-

receive input image

convert RGB to gray scale

background subtraction

threshold method

mathematical; morphology method

vehicle classification

region of interest for motorcycle

helmet detection
 IV. IMPLEMENTATION

The main objective of this system to develop a system which detects whether the person is wearing helmet or not and then
capture the image of the number plate and generates fines on the riders. Here we are determining it by using the Kstar algorithm.
The figure explains about the real time vision based monitoring system for motorcycle to detect and track motorcyclists who
break the rules. The proposed system aims to catch the traffic rule violators who are not wearing helmets and report the vehicle
number plate violators and generate fines.

1. Image procurement: A camera is used to capture vehicle on road in traffic. It is very first part of any vision system.

2. Preliminary processing techniques: In this step mainly focus on removal of background noise, enhancing of contrast
[7] and binarization of images.

3. Vehicle classification: By considering two parameters (Aspect ratio and area) of particular vehicle, Motorcycles are
classified and processed further.

4. Helmet detection: After extracting head part (ROI), [11] classifier which is being trained by certain amount of pictures
of helmets is provided with features of ROI. By matching ROI and trained features, it will be determined that whether
motorcyclists is wearing helmet or not.

Flowchart :-
 V. RESULT & DISCUSSION

We have divided the experiment into 2 parts. One is detection of vehicle and another one is detection of helmet. The vehicle
detection result is shown in Table 1, and the result of the helmet detection is shown in Table 2.

Table 1: Result of vehicle detection

No. of Actual Vehicles No. of detected vehicles Accuracy

25 23 92 %

Table 2: Result of helmet detection

Motor cyclist No. Correct detection Accuracy

With helmet 15 13 86.66 %

Without helmet 10 8 80 %

Fig 2. The detection of motorcyclist and helmet

VI. CONCLUSION

This system gives an idea about the number of traffic offenders in an area. It generates a database of all bike riders driving
without wearing a helmet along a snapshot for proof. Use of open and free technologies like tensorflow, OpenCV and tesseract
makes the software relatively less expensive. Under fair lighting conditions, this system was tested to give full proof and accurate
results. The overall awareness to public will increase the impact of the system.
 VII. FUTURE ENHANCEMENT

This project can be further improved by implementing advanced safety measures like to check collision detection, capturing
images of vehicles with who breaks the rule by riding with high speed, capturing images of drivers while talking on phone and
driving. This lessens the work of traffic police and with affordable cost. Once installed correctly, this system will work and
generate databases for greater period of time under proper maintenance.

VIII. REFERENCES

[1] A. Leelasantitham and W. Wongseree, “Detection and classification of moving thai vehicles based on traffic engineering
knowledge,” in ITST 2008, oct. 2008, pp. 439 –442.

[2] J.Chiverton ”Helmet presence classification with motorcycle detection and tracking in IET Intelligent Transport Systems”,
vol.6, Issue 3, IEEE 2012.
[3] “Automatic detection of Motorcyclists without helmet” by Silva, Kelson Aires, Thiago Santos, Kalyf Abdala, Rodrigo Veras,
Andre Soared, Departmento de Computacao,2013,Latin America Computing Conference, IEEE 2013.
[4] C.-Y. Wen, S.-H. Chiu, J.-J. Liaw, and C.-P. Lu, The safety helmet detection for atm’s surveillance system via the modified
hough transform, in IEEE 37th Annual International Carnahan Conference on Security Technology., 2003, pp. 364–369.

[5] “Detection of motorcyclist without helmet using video convolution neural network” by C. Vishnu, Dinesh Singh, C.Krishna
Mohan, Sohan Babu; International joint conference on Neural Networks(IJCNN),IEEE 2017.
[6] Amit Verma et al.(2010), “Vehicle Number Plate Detection using Sobel Edge Detection Technique” IJCST, Vol. 1,Issue 2.

[7] “Automatic License plate recognition using OpenCV and neural network ”by Sweta kumar., Leeza gupta, Prena gupta,vol-
5,issue-3,may-june 2017.

[8] M. Zengqiang, P. Cunzhi, H. Ke, and C. Qiandong, Zengjia Yan, Research on segmentation of overlapped vehicles based on
feature points on contour, in FBIE 2009, dec. 2009, pp. 552 –555.

[9] K. Takahashi, Y. Kuriya, and T. Morie, Knoeppel, Bicycle detection using pedaling movement by spatiotemporal gabor
filtering, in TENCON 2010, nov. 2010, pp. 918 –922.

[10] S. Sonoda, J. K. Tan, H. Kim, S. Ishikawa, and T. Morie, Moving objects detection at an intersection by sequential
background extraction, in ICCAS 2011, oct. 2011, pp. 1752 –1755.

[11] S. G. Mallat, “A theory for multiresolution signal decomposition: the wavelet representation,” Pattern Analysis and Machine
Intelligence, IEEE Transactions on, vol. 11, no. 7, pp. 674–693, 1989.