Multiple object tracking for fall detection in real-time surveillance system

The algorithm using 3D spatio-temporal templates is applied to occlusion problems for detecting and tracking people accurately and shows robust multiple object tracking and a good detection rate of falls in real-time.

Abstract

In this paper, we present multiple object tracking to detect falls using a low-cost single and uncalibrated camera in real-time environment. Until now, existing studies for fall detection only presented their methods for a single person tracking. Occlusion problem is one of the main challenges in health care surveillance systems. For occlusion handling, 2D modeling tracking to multiple object tracking is not enough. The algorithm using 3D spatio-temporal templates is applied to occlusion problems for detecting and tracking people accurately. We use 2D trajectory information obtained in order to distinguish fall activities from normal daily activities. The experimental results show robust multiple object tracking and a good detection rate of falls in real-time.