OBJECT TRACKING USING BRANCHING PARTICLE FILTERS

Abstract

In digital video communication systems it is important that a video to be compressed, because of storing capacities as well as bit-rate constraints. The video processing is done using Sum of absolute Differences and with the image processing block set. We first calculate motion vectors between successive frames and use them to reduce redundant information. The method uses optical flow algorithm to calculate changes in the intensity of the pixels of the images. These apparent velocity components are then subjected to various image processing techniques to obtain centroid of the vehicle across the object. Here we are interested in finding spatio-temporal interest points of our object having a significant local variation of intensities of image. Branching particle filtering is a novel technique which is used to reduce the error in and computation time in tracking algorithm and is applied on motion frames to show these results for object detection. The distance traveled by the vehicle is calculated using the movement of the centroid over the frames. The image coordinates of the centroid are mapped to World space. The world coordinate distance is further mapped to actual distance using pixel to distance ratio. Using this information the velocity of the vehicle is estimated. In this Thesis we developed object tracking for real time video which, demonstrates the motion compensated video processing by using key point descriptor. First we have taken an object as reference object or image then the next successive object is compared with the reference object or image. Each time the successive object is compared with the reference object and produces an absolute difference then the summation of all these differences shows its sum of absolute difference.