Human motion capture data have been widely used in computer animation for creating realistic human motions. Other research fields such as sports medicine, physical therapy, also benefit by gaining insights of various motion characteristics from analyzing captured human motion data. As huge collections of motion data are rapidly becoming available, there is an immediate need for tools to efficiently identify and extract suitable motions at various levels of abstractions. In this paper, we propose a novel method that supports retrieving motion sequences based on similarity in behaviors. We treat a complex motion sequence as a concatenation of segments of distinct behaviors. We characterize each motion segment by a set of features that describes the distribution of the poses in the segment, and then use these features as the modeling primitives to construct a compact but effective indexing scheme. Our method is robust to spatio-temporal variations among similar motions. We demonstrate the strength and efficiency of our method with a large human motion database.
Guodong Liu, Leonard McMillan. Retrieval of Motion Capture Data Based on Behavioral Similarity. In Posters at the ACM SIGGRAPH/Eurographics Symposium on Computer Animation (SCA 2006), Vienna, September 2006.
2-Page Poster in PDF format (404 Kb)