REAL TIME OBJECT TRACKING IN COLOR VIDEO WITH AUTOMATED TARGET SELECTION

Abstract

Object tracking is an essential task in application domains such as traffic monitoring, automated surveillance and robot navigation. Tracking objects can be challenging due to object fading, noise and background clutter, varying appearance, orientation, scale, and velocity of the maneuvering object and object occlusion. The work undertaken in this dissertation is mainly focused on development of a reliable and robust tracking system that can track any object of interest in a video acquired from a stationary or moving camera. A procedure for automatic target selection has also been proposed and implemented that can select target with the help of single click at the object of interest in a single frame. The proposed automatic target selection algorithm is based on segmentation technique and extracts a bounded object (hole) around the point selected by user. The steps involved in the algorithm are smoothing, edge enhancement and morphological operations. The proposed visual tracking system has been implemented in the RGB space. To reduce the computational complexity of the system, RGB histogram based feature set has been extracted. Bhattacharya coefficient is used as similarity measure between template and each template size section of the search window. The varying appearance of the object and the short-term neighboring clutter is addressed by dynamic template updating scheme. This scheme augments the feature set values of template iteratively, by taking weighted sum of template and the current best-match. This scheme minimizes the template drift phenomenon and copes with transient occlusion and background clutter. The search for the target is carried out in a dynamically generated resizable search-window instead of whole frame to reduce computational complexity and false positive. The size of the search window is determined based on the change in position in previous iteration. To cope with the target scaling challenge, an adaptive scale adaptation algorithm has been implemented in which the size of the section at the best-match location is varied and RGB histogram based features of this varied size section is compared with the RGB histogram based features of target. The proposed system was tested on videos with diverse objects with above mentioned issues. The results show that the proposed system can handle most of the issues.