Appearance Model Based Single Object Tracking

Visual object tracking is one of the fundamental problems in computer vision. It refers to automatically estimate the state and trajectory of an identified object in video sequence. It has a wide range of applications in surveillance, intelligent traffic, human computer interaction and video compression, etc. Note that the above applications heavily rely on the information provided by the trackers. Although significant process has been made in recent years, visual tracking remains a challenge in practical applications due to numerous factors such as pose variation, motion blur, heavy occlusion, illumination change, and background distraction. These factors will gradually degrade the robustness of the tracker and may finally lead to the tracking failure. Therefore, the study on robust visual tracking algorithms is still very important.We focus on appearance model based single object tracking in this thesis. The main contributions are as follows:(1) A robust visual tracking algorithm based on joint sparse dictionary learning;(2) A robust visual tracking algorithm based on local kernelized representation;(3) A robust visual tracking algorithm based on online deep manifold learning. Experiment results demonstrated the effectiveness and efficiency of the proposed methods.