一种基于Hausdorff距离的运动物体跟踪算法

针对视频处理中运动物体的检测和跟踪问题,提出了一种基于Hausdorff距离的目标跟踪算法。新算法提出首先采用多尺度分水岭变换获取运动物体模型,消除了传统基于分水岭变换算法存在的缺陷;然后使用部分Hausdorff距离实现后续帧中运动物体模型的匹配;最后再次使用多尺度分水岭算法完成运动物体模型的更新。实验表明,该算法可以有效地跟踪多个刚体或非刚体目标。

Moving object tracking algorithm based on Hausdorff distance

It is a hard job to track the moving objects in video sequence. Considering disadvantages of watershed transform, a novel moving object tracking algorithm was proposed based on Hausdorff distance. In the algorithm, firstly Canny edge detector was adopted for the image boundary generation, and multi-scale watershed transform was used to initialize contour of the objects. Then the matching of the objects was judged by partial Hausdorff distance. Finally, multi - scale watershed transform was reused to update the object model. Experiment results show that the proposed algorithm can efficiently track more than one non-rigid moving object simultaneously.