Vehicle Identification Using Iamge Processing Techniques

Abstract

This thesis work is focus on detection and recognition of moving vehicle from a recorded video which is captured by camera from real life traffic scenario. A Frame subtraction algorithm has been used to identify the moving object but this method is sensitive to environment changes. To overcome this limitation, some other algorithms along with this has been used that can help to generate traffic data even at night and under bad weather condition. The proposed methodology assumes a static camera and enough light condition to detect the object. The object recognition is done on the basis of similarity of boundary signatures of objects. Boundary signatures are surface feature vectors that reflect the probability of occurrence of a feature of a surface (or an object) boundary. Here, the boundary signature is used as the shape of the object. The measurement of similarity is preceded by solving the correspondence between boundary points on the two shapes and use of correspondence to recognize the object. In order to solve the correspondence problem, a descriptor designated as shape context, is attached to each point. The shape context at the reference point captures the distribution of remaining points relative to it. Corresponding points on two similar shapes will have same shape contexts. The dissimilarity between two shapes is computed as the sum of matching error between corresponding points. The object is recognized on the basis of least dissimilarity with respect to the stored prototype shapes. To remove the problem of alignment, the surveillance camera is placed on the side of the road and the entire vehicle will go perpendicularly to the view of camera. This algorithm can categorize the object as heavy or light weight vehicle and further light weight vehicle in jeep, car and two-wheeler with sufficient accuracy to develop practically useful application.