Intelligent Video Surveillance: Behavioral Detection of Falls Based on Double-Layer Support Vector Machine

Abstract

The elderly population is expected to more than triple by 2050 in the United States alone. This indicates the growing need of medical innovations that are expected to deliver convenient, user-friendly, and intelligent health care in the home. In particular, the use of computer vision and artificial intelligence offers a new promising solution to analyze human behavior and detect unusual events. We propose a novel method to detect unintentional falls, which are one of the greatest risks for seniors living alone. Our approach is based on a machine learning technique, in which we use a human skeletal joint data from a Red Green Blue Depth (RGBD) sensor, Kinect. To the best of our knowledge, this is the first fall detection research utilizing machine learning mechanism with joint information. After building a solid definition of various types of falls, the system is trained using Support Vector Machine (SVM) in two separate layers, which we call Double Layer SVM (DLSVM). Our evaluation results show that our proposed system can efficiently detect different types of falls facing various directions from the camera and is capable of accurately distinguishing an actual fall versus a fall-like behavior. ⓒ 2013 DGIST