Real-time human motion detection and classification

Tracking the movements of different body parts of humans and classifying them in certain states is an important problem in human-computer interaction. In this paper we propose a robust mechanism for the tracking and classification of movements of hands and legs of a person in real-time. The technique employs a localized and adaptive frame-based motion detection mechanism that not only caters for considerable intensity variations in the video streams, but also takes care of deformation in image, rotation in the human body and self-occlusion of body parts. To make system work in real-time, a person wears different color bands on his/her hands and legs. The simulation built on this mechanism continuously processes the video stream, detects the movements of hands and legs and classifies them in certain states, which can be labeled as {left/right} hand punched, and {left/right} hand unpunched {left/right} leg kicked, {left/right} leg unkicked.