基于事件分析的多摄像机鲁棒跟踪算法

提出了一种新的基于事件分析的目标跟踪算法来解决多个目标分离或遮挡时的可靠跟踪问题.首先提出使用仿射变换来获得多个摄像机之间重叠画面的映射关系,实现目标交接,为后面的目标识别奠定基础.然后当单摄像机目标跟踪过程中发生候选目标多于一个或者多个目标对应一个候选目标的情况时,提出一种判别目标出现遮挡事件或分离事件的新方法,并且通过多摄像机的目标交接准确识别出发生遮挡或分离事件的目标标号,解决目标发生遮挡或分离后跟踪失败的问题.实验结果证明:所提出的方法突破了一般跟踪算法受目标底层特征约束的难点,具有更高的鲁棒性.

Robust video tracking based on event analysis

Reliable video tracking under occlusion is a challenging task. A novel tracking method based on event analysis and multi-camera communication is proposed. Firstly, the relationship between the overlay parts of two cameras is computed for object handoff using affine transform. Then, a new algorithm for identifying Occlusion Event (OE) and Separation Event (SE) is designed. When multiple targets are tracked, or there are more than one tracked candidates in the case of single camera tracking, the proposed algorithm recognizes the IDs of tracked candidates by target handoff among multiple cameras, resulting successful tracking. The experimental results show the robustness of the proposed method without using any low-level features of the tracked objects during the tracking process.