Object tracking is one of the fundamental topics of computer vision with diverse applications. The arising challenges in tracking, i.e., cluttered scenes, occlusion, complex motion, and illumination variations have motivated utilization of depth information from 3D sensors. However, current 3D trackers are not applicable to unconstrained environments without a priori knowledge. As an important object detection module in tracking, segmentation subdivides an image into its constituent regions. Nevertheless, the existing range segmentation methods in literature are difficult to implement in real-time due to their slow performance. In this thesis, a 3D object tracking method based on adaptive depth segmentation and particle filtering is presented. In this approach, the segmentation method as the bottom-up process is combined with the particle filter as the top-down process to achieve efficient tracking results under challenging circumstances. The experimental results demonstrate the efficiency, as well as robustness of the tracking algorithm utilizing real-world range information.
展开▼
