基于多摄像头监控的人体跌倒检测算法  

Human Fall Detection Algorithm Based on Multi-Camera Surveillance

作　　者：魏振钢[1] 孔勇强 魏兆强[1] 张小龙[1] WEI Zhen-Gang;KONG Yong-Qiang;WEI Zhao-Qiang;ZHANG Xiao-Long(College of Informatiou Science aud Engineering , Oceau University of China , Qingdao 266100, China)

机构地区：[1]中国海洋大学信息科学与工程学院,山东青岛266100

出　　处：《中国海洋大学学报（自然科学版）》2019年第7期142-148,共7页Periodical of Ocean University of China

基　　金：国家体育总局科技服务项目资助~~

摘　　要：人口老龄化使得空巢老人数量越来越多。跌倒作为威胁独居老人生命安全的主要因素受到了社会的广泛关注。为保护老人的生命健康不受威胁,本文提出一种基于计算机视觉的两级人体跌倒检测算法。从监控摄像机中采集视频数据,对其做前景提取,通过形态学操作为前景块绘制矩形边界,根据矩形宽高比从中筛选出所有可能是跌倒的行为,这是粗粒度级检测。之后再用统计学方法对第一级检测出的前景块绘制椭圆边界,分析其形态变化,最终检测出跌倒行为,这是细粒度级检测。本文在开源多摄像头跌倒数据集上进行了评估。仿真实验和与当前的先进方法的对比表明本文算法取得了非常好的效果。The number of empty-nesters is continuously increasing due to ageing of population. As the main factor that threatens the life of empty-nesters, falls have received wide attention in society. This paper presents a dual-stage fall detection algorithm based on computer vision, which aims at protecting the life and health of elderly. Background subtraction is used to extract foreground image from surveillance video. We first draw a bounding box for the foreground blob by morphological operation, and recognize all potential falls using aspect ratio of the bounding box(coarse granularity level). After that, we draw ellipse for the foreground image of potential falls based on statistical method, and analyze variations of ellipse to achieve fall detection(fine granularity level). This paper have been conducted on a publicly available multi-camera fall dataset. Simulation experiments and comparison with state-of-the-art show the effectiveness of the proposed algorithm.

关 键 词：视频监控 跌倒检测 前景提取 形态分析 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]