Fall detection using k-nearest neighbor classification for patient monitoring

The incident of fall of elder people increases day by day. Falls are one of the greatest risks for seniors living alone. Sometimes people may get serious injury to the spinal cord and hip region. In such cases, an injured elder people may remain on the ground for several hours after a fall incident has occurred. So there is a need of fall detection system to avoid such incident. This paper propose a novel method to detect falls which combines four features, Orientation angle, ratio of fitted ellipse, Motion Coefficient, Silhouette threshold. These features act as inputs to K-Nearest Neighbor classifier which recognizes fall events. This algorithm gives accuracy above 95% on stored video sequences of activities and real time environment.