Vehicle Identification
and Classification System
Name : Komal Vishnu
Id:2300032086
DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING
Contents :
• Introduction
• Problem Statement
• Applications
• Vehicle Recognition Process
• References
Introduction
Introduction
• The VICS system for identification and classification of moving vehicles on
the road side from the videos is a great importance today.
• It is an active research area in Intelligent Transportation Systems (ITS).
• Vehicle recognition has been an area of recent interest to the Computer
Vision community, no prior research study has been used to build an onroad
vehiclerecognition.
Computer Vision
Computer vision is an important field of artificial intelligence where
decision about real worldscene having high dimensional data is taken.
Cont.
Applications of Computer Vision
Problem Statement
Problem Statement
• The main goal of our project is to implement an efficient method for
recognizing vehicles in Indian conditions. There are some challenges
in these conditions that makes vehicle recognition much harder.
1. In India, most of the cases traffic system is non lane based.
2. Road conditions are more varied and traffic is unstructured.
3. Vehicles are parked frequently by the sides of the roads.
4. Within same vehicle class there are large variety’s and models. These
looks different in size and appearance. It is generally observed in Indian
vehicles like cars and Truck’s.
5. Shapes of the vehicles have a key role in recognition, there is high
intraclass variance among Indian vehicles.
Challenges
• Vehicle recognition task deals with only images in outdooror natural
lighting it is known to cause noise related problems.
• The other is typical geometry of vehicles. Vehicleschassis are
constructed in a significantly greater variety ofgeometries as compared
to other objects. There are number of dissimilarities in vehicles like
height, number of wheels, body shape and color.
Sample Images from Indian Vehicle database
includes four classes (Truck, Auto, Bus and Car respectively).
Images were captured with variance in pose, view and
lightningconstraints.
Applications
Applications
• Electronictoll collection management.
• Identification of unauthorized vehicleson roads as a part of
vehiclesurveillance and traffic data analysis.
• License plate localization.
• Computer assisted driving.
• Methods for reducing road accidents.
Traffic Management System
• Thetraffic on the roads is increasing day by day. There is dire need
of developing an automation system that can effectively
manage and control the traffic on roads.
Vehicle Recognition Process
Vehicle Recognition Process
• Vehicle recognition process in Indian scenario has several challenges.
To address these challenges features of an individual vehicle from
different directions are to be considered.
• We are using Bag of features (BOF) and Support vector machines for
vehicle recognition.
Process
• Bag of Features (BOF) works on the principle that every object can be
represented by its parts. For example, a Truck contains parts like bigtyres,
number plate, cabin etc. Also car contains wheels, number plate and
windows, but the basic difference between an Truck and car is
observed to be in size and tyres. So, in order to recognize an object it is
necessary to first recognize the parts of it and based on the parts identify
the object correctly.
• Basing on above principle vehicle recognition process is initiated to
recognize parts of the vehicle images by extracting image patches. This
can be done by the combination of Harris corner and Sift points
because these contain rich local informationof the image.
Cont.
• For each patch, descriptors were calculated in the form of vectors. To
address the cardinal and ordering problems, clusters are generated by
apply K-means algorithm on these descriptors. Each cluster is refereed
as word for an image. Thus image is represented as a bag of words or
bag of visual words. These words are representedas a histogram and
referring as histogram of features. Features obtained in the above
process are classified by train and test process using SVM.
Reasons of using SVM
• The reason for the select of SVM as a classifier is that, it has some superiority
over other approaches. The important points are global minimum
solution, learning and generalization in huge dimensional input spaces,
use of kernel function and classification is done by the separating hyper
plane at a maximum distance to the closest points in the training set.
Challenges
• Objects in small size
• Novel model
• Clutter
• Poor quality images
References
• R.S Vaddi, L.N.P Boggavarapu, K.R Anne, "Computer Vision based
Vehicle
Recognition on Indian Roads", International
Journal Of ComputerVisionAnd Signal Processing, 5(1), 8-13(2015)
• Baljit Singh Mokha and Satish Kumar, A Review Of Computer
Vision System For The Vehicle Identification And Classification From
Online And Offline Videos.
Thank You
