Gesture segmentation in complex motion sequences

摘要: Complex human motion sequences (such as dances) are typically analyzed by segmenting them into shorter sequences, called gestures. However, this segmentation process is subjective, and varies considerably from one observer to another. In paper, we propose an algorithm hierarchical activity segmentation. This employs a dynamic layered structure represent the anatomy, uses low-level parameters characterize in various layers of hierarchy, which correspond different segments body. characterization used with naive Bayesian classifier derive creator profiles empirical data. Then those predict how creators will segment gestures other sequences. When predictions were tested library 3D capture segmented 2 choreographers they found be reasonably accurate.