A view invariant gait cycle segmentation for ambient monitoring

Gait analysis has many clinical applications in disease detection and treatment evaluation. Gait cycle segmentation is a critical component in gait analysis for timing the gait phases in evaluating many movement disorders. Computer vision techniques have been widely used in surveillance for security monitoring. They are nonintrusive and do not require cooperation from subjects. In this paper, we propose to leverage the videos from existing surveillance monitoring systems to provide long-term and ambient assessments of gait patterns from subjects' daily activity without the requirement of wearing a device. Our proposed method is a novel view-independent method for gait cycle segmentation. We use the temporal duration of spatial features to achieve fast, robust and accurate gait cycle segmentation. The method take videos from a single non-calibrated camera and is not limited by specific viewing angles of the subject.