3D measures computed in monocular camera system for fall detection

Abstract

Traumas resulting from falls have been reported as the second most common cause of death. For this reason, computer vision tools can be exploited for detecting humans’ fall incidents. In this paper, we propose a fast, real-time computer vision algorithm capable to detect humans’ falls in complex dynamically changing conditions, by exploiting the motion information in the scene and 3D space’s measures. This algorithm is using a single monocular low cost camera and it requires minimal computational cost and minimal memory requirements that make it suitable for large scale implementations in clinical institutes and home environments. The proposed scheme was tested in complex and dynamically changing visual conditions and as proved by the experiments it has the capability to detect over 92% of fall incidents.