Watershed region tracking for video motion description
Takashi Kobayashi, Xiaolin Zhang, Motoyuki Sato · 2006
A video motion description method that uses watershed segmentation and graph representation is presented. In our approach, a video sequence is divided into watershed regions in each frame, and extrema of the regions are employed for describing video motion. Watershed region graph is temporal relation of the regions that are represented as directed graph, where nodes are the regions, and directed edges are links of the regions. The links connect the regions between two adjacent frames and structural pattern of the graph represents region behavior that consists of transit, merge and split. The presented technique is applied to some video sequences and the results illustrate that the technique is capable of describing video motion with occlusion.