Ultra-wideband radar and vision based human motion classification for assisted living

Fall detection for elderly is one of the most important areas in elderly healthcare. Both video and radar based detections are being developed for this purpose. This paper presents a new approach to classify different human motions through machine learning. In particular, our objective is to achieve high-accuracy fall detection through the exploitation of both video and radar data. Motion history image is applied to extract temporal features from video clips, and hidden Markov models are trained with the features extracted from both video and radar data to discern the types of motion. Experiment results indicate that the proposed approach provides improved performance in distinguishing falls from other motions such as sitting.