Object tracking is to detect, localize and track moving object in video sequences captured by cameras. Object tracking plays an important role in visual analysis and understanding of human motion, it's the intermediate-level vision part. By visual tracking, we obtain the motion parameters, the poses and the trajectories of objects, which lays a solid foundation for the high-level activity understanding and recognition. Although object tracking is really a hot research topic in recent years, there do exist many theoretical and technical problems, especially in the cases of noise, motion blurring, illumination changes, occlusion etc. Trajectories in video contain sufficient spatial and temporal information. They represent not only the spatial location of the object, but also the velocity and acceleration. Combined with the scene information, trajectory can also represent the high-level semantic information. Trajectory pattern learning is part of high-level object motion analysis. It is simply viewed as an unsupervised trajectory learning problem. After trajectory learning, several trajectory patterns are obtained. Therefore, the core of trajectory pattern learning and retrieval is how to learn and model trajectory patterns in an effective and efficient way. This thesis mainly focuses on robust object tracking and trajectory pattern learning. Specifically, we mainly discuss two topics: (1) appearance modeling based object tracking; (2) object trajectory pattern learning. Experimental results have demonstrated the effectiveness and robustness of our methods. The main contributions of this thesis are summarized as follows: 1. We propose a pyramid based appearance modeling for robust object tracking. Based on the incremental subspace learning, we introduce the idea of multi-scale analysis and divide the target appearance into three-level pyramid. Using incremental PCA, we conduct subspace appearance modeling for every sub-block of each pyramidal level and calculate their reconstruction errors respectively. In the particle filter framework, we conduct posteriori state estimation for the target and get tracking result. In addition to the pixel value feature, we make full use of the pyramidal structure and add the Haar-like feature and PHOG feature. They can describe the texture information and shape information of the target, making tracking result better. 2. We propose a trajectory learning framework using graph-based semi-supervised transductive learning...
修改评论