TR#487: Recognizing Movement using Motion Histograms

James W. Davis

In this paper, we present a real-time computer vision approach to recognizing human movements based on patterns of motion. The underlying representation is a Motion History Image (MHI) which is characterized by multiple histograms of the local motion orientations. The approach is adapted to accommodate movements of different durations by using a simple iterative technique. Quantitative results are given showing discrimination between different human movements using the approach. An extension addressing occlusion and distractor motion is also presented within this framework.