Action Recognition.pdf

Action Recognition based on Human Movement Characteristics? Radu Dondera, David Doermann and Larry Davis University of Maryland College Park, MD USA {rdondera, lsd@}, {doermann@} Abstract We present a motion descriptor for human action recog- nition where appearance and shape information are unre- liable. Unlike other motion-based approaches, we lever- age image characteristics specific to human movement to achieve better robustness and lower computational cost. Drawing on recent work on motion recognition with bal- listic dynamics, an action is modeled as a series of short correlated linear movements and represented with a proba- bility density function over motion vector data. We are tar- geting common human actions composed of ballistic move- ments, and our descriptor can handle both short actions (e.g. reaching with the hand) and long actions with events at relatively stable time offsets (e.g. walking). The pro- posed descriptor is used for both classification and detec- tion of action instances, in a nearest-neighbor framework. We evaluate the descriptor on the KTH action database and obtain a recognition rate of 90% in a relevant test setting, comparable to the state-of-the-art approaches that use other cues in addition to motion. We also acquired a database of actions with slight occlusion and a human actor manipulating objects of various shapes and appearances. This database makes the use of appearance and shape in- formation problematic, but we obtain a recognition rate of 95%. Our work demonstrates that human movement has distinctive patterns, and that these patterns can be used ef- fectively for action recognition. 1. Introduction Human action recognition is an active field of computer vision research with applications to visual surveillance, hu- man computer interaction and video indexing. One of the main challenges in action recognition is action representa- tion. Previous work has investigated the use of appearance, shape, motion and sequencing informa