Study On Object Tracking With Geometrical Appearance Change

Object tracking is a locating process of moving object, in consecutive frames. As a research focus in computer vision, object tracking has gained rapid development in the past far years. In many areas, such as traffic monitoring, video communication, and robot vision, object tracking has been widely used.In recent years, various tracking algorithms have been put forward, and have obtained considerable results. But for non-rigid objects with structural change, especially those with geometrical appearance change, object tracking is still a huge challenge in computer vision. Hence, to solve object tracking problem with geometrical appearance change is a looming problem in object tracking area.In this paper, we focus on the tracking problem of geometrical appearance changing object. We studied current related object tracking methods. According to our analysis, the multiple features and part-based model based geometrical appearance changing object tracking method has been put forward.In our problem, the selection of object features has great impact on the tracking results. Just depending on global features can’t adequately describe the object with geometrical appearance change. For this reason, we select multiple features for the part based model to describe the geometrical appearance changing object. Part based local models are established, and multiple features are extracted to describe the target. During the process of object tracking, in order to adapt to the geometrical appearance change, the model needs to be updated.After object modeling and multiple features extraction, the multiple features need to be classified. In this paper, we adopt multiple weak classifiers integrated into a strong classifier to differentiate the target from background, so as to improve the accuracy of classification and tracking. In the process of tracking, just like part based model, the classifier also need to be updated.By comparing multiple experimental results, we verified the effectiveness of our multiple features and part based model based geometric appearance changing object tracking method.We also demonstrated the strong adaptability and superior performance of our method for geometric appearance changing object tracking.