Motion Retrieval Using Probability Graph Model

In this paper, we propose a content-based motion retrieval algorithm. In this work, firstly, each motion is represented by a set of sequence frames. Representative frames are first selected from the motions and the corresponding weights are provided. Secondly, the graph model is built with these selected frames. For searching the optimal measurement between query motion and relevant motions in database, an object function is built. The task to find the maximal a posterior (MAP) in the motion level is equivalent to find the minimal objective function value. At last, based on probability calculation, the KM (Kuhn-Munkres) algorithm is used to find the optimal matching between motions. The matching result is used to measure the similarity between two motions. Experimental results and comparison with existing methods show the effectiveness of the proposed algorithm.