Motion descriptor based on Aligned Cluster Analysis

The data of human motion is complex both spatially and temporally, which makes it a difficult work to extract motion features in an efficient way. In this paper, we propose a novel descriptor for human motion based on Aligned Cluster Analysis (ACA) and then this descriptor is used in retrieval. Firstly, Aligned Cluster Analysis (ACA), a robust method used to segment sequence of motion, is taken to capture data into actions. Secondly, body joints are denoted by quaternion, each sub-segment of motion features is extracted by k-means clustering. Finally, the motion features match with the features database. Experimental results prove that the proposed approach can accurately segment the motional sequence. The motional features are successfully employed in the retrievable process.