| Object tracking is an important field of computer vision research and has broad application prospects. The main contents of this paper focus on two aspects: for single-target tracking, we propose a part-based tracking with appearance learning and structural constrain method; for multi-target tracking, a novel tracking method based on detection is proposed.A class method of tracking based on online learning the appearance model of the object has shown good results. But deformable targets and partial occlusions continue to represent key problem in this area. The method we proposed can well handle these problems. First, we take advantage of the existing online learning appearance model to learning the appearance of each part. Second, we propose a novel part initialization method and an affine invariant structural constrain between these parts. Third, a tracking model based on the appearance of each part and the spatial relationship between the parts is proposed.In the field of pedestrian tracking, we need to take the detection results as the initial state of object tracking. It is challenging task to develop a robust tracking method, due to factors such as false positive, missing detections and occlusion. Greedy data association method is used to decide which detection should guide which tracker. When a matching detection is found, the position and the size of the detection rectangle will robustly guide the particles. We demonstrate the performance of our algorithm through experiments. Finally, our method is applied to real-life scenarios successfully. |
