一种视频序列图像中区域匹配方法

A Region Matching Algorithm for Video Sequence

提出一种非常有效的区域匹配算法,能够通过在相邻视频帧中分割出自然形状的区域后,进行相关匹配。在静止背景或背景变化较小条件下,实现目标与背景的分离。分割采用分水岭算法和利用时、空域信息的小区域合并技术,最后利用经改进的颜色直方图分析法匹配区域。实验结果表明,尽管区域对象会发生一定形变,依旧能获得较理想的效果。该方法可用于视频序列中对目标的检测与跟踪。

We propose an effective region matching algorithm for further object tracking purpose in this paper. In order to implement the matching, it is of key importance to generate natural shape regions in the video frame. The generating process can be fulfilled by two steps sequentially. Firstly, watershed segmentation algorithm is adopted based on the calculated gradient image. Then small regions are merged by our proposed spatial-temporal analysis method. The advantage of the merging is that the boundary between the moving objects and the background can be clearly divided. After doing all these work, region matching was applied according to our improved color diagram analysis approach so as to match related regions between two neighbor frames. This proposed method is especially applicable to the cases when the video′s background is changeless or slightly-changed in our test. It is proved that our method is robust, and many experimental results show that satisfactory segmentation and matching effect are obtained even with some deformation in the relative regions. Since our method can provide the objects′ moving information, it is very useful for performing object tracking in video sequences.