Spatial temporal pyramid matching using temporal sparse representation for human motion retrieval

An efficient retrieval mechanism is essential to search for a particular motion from a large corpus. This has proven to be a challenging task as human motion is high dimensional in both spatial and temporal domains. Besides, semantically similar motions are not necessary numerically similar because of the speed variations. In this paper, we propose a temporal sparse representation (TSR) for human motion retrieval. Compared with existing methods that adopt sparse representation, our TSR encodes the temporal information within motions and thus generates a more compact and discriminative representation. In addition, we propose a spatial temporal pyramid matching kernel based on TSR, which can be used for logical comparison between motions. Moreover, it improves the effectiveness of motion retrieval in terms of accuracy and speed. Through our experimental evaluations, we demonstrate that the proposed human motion retrieval system has better performance and allows the user to retrieve desired motions from the motion capture database.