Object Tracking Based On Sparse Representation And Features Selection

Object tracking is an important task within the field of computer vision. The goal of tracking can be defined as the problem of estimating the object-centric information in each frame, such as positon. shape, size and rotation. With the development of image processing technology and proliferation of high-powered computers, there are many students, security department and researchers who have strong intresting in object tracking. The use of object tracking is pertinent in many applications such as motion-based recognition, automated surveillance, video indexing, human-computer interaction, vehicle navigation and traffic monitoring.Tracking objects can be complex due to some challenges are caused by the presence of noise in image, partial and full object occlusions. varying viewpoints of object, background clutter, illumination changes and real-time processing requirements. With the researching of various countries’ scholars and technicians continuously, numerous approaches for tracking have been proposed to deal with these changes in the last decade.In order to tracking object in complex scenes, the paper proposes an online features selection mechanism based on sparse representation in the Lucas-Kanade image registration framework. To reduce the impact of interference on the tracking, the features that best discrimination betweem object and adjacent background are selected in each frame of the video sequence. The algorithm is composed of forward dictionaries and background dictionaries, the former which will be created manually from the first frame and updated with the tracking results, the background dictionary will be reconstruct by evrey frame. Meanwhile, a new dictionary updating strategy not only can effectively cope with the appearance changes, but also handle drift. Experiment shows that the proposed algorithm can effectively deal with pose change, illumination change and large partial occlusion.