基于KinectV2的跌倒行为检测与分析

2019-04-04李文阳马行穆春阳

现代电子技术 2019年6期

李文阳马行穆春阳

关键词： Kinect V2; RGB?D; 骨骼图像; 阈值分析; 意外跌倒; 实时检测

中图分类号： TN911.23?34 文献标识码： A 文章编号： 1004?373X（2019）06?0142?04

Abstract： The Kinect V2 somatosensory equipment is used to detect and identify the fall behavior of the elderly， so as to reduce the casualties caused by falls of the elderly as much as possible. The human body skeleton image and position information are obtained by processing the RGB?D images acquired by the Kinect camera. The skeleton tracking technology is used to select skeleton points such as the center point of the human body， center point of two hips， and right foot sole， so as to calculate the parameters such as the center point spatial position of the human body， body movement speed， position of the center point of two hips， and height of the center point of two hips from the ground in real time. The speed threshold and height threshold during the falling are emphatically analyzed in this paper. A large number of verification experiments were carried out. The results show that the algorithm proposed in this paper has a high real?time performance in the indoor environment， can overcome the problems of low detection rate and poor real?time performance of the traditional video detection technology， effectively protect the privacy of the elderly during the detection process， and perform real?time detection continuously， which can provide security assurance for accidental falls of the elderly.

Keywords： Kinect V2; RGB?D; skeleton image; threshold analysis; accidental fall; real?time detection

0 引言

随着我国人口老龄化进程的加速，根据数据显示预计到2050年，中国老龄化人口将达到4.5亿人，约占总人口[1]的[13]。然而跌倒已成为我国伤害死亡的第四位原因，在65岁以上的老年人中占首位，并且死亡率随着年龄增加而急剧上升，此外还可能导致大量残疾，影响身心健康。根据Noury等人的研究[2]，如果老年人在跌倒后能得到有效的救助，可以有效降低老年人死亡的风险，因此在室内对老年人跌倒检测是一项具有实际意义的事情。传统的对老人跌倒检测方法有以下3种方法：

1）利用穿戴式传感器来检测人体跌倒[3];

2）视频图像分析法[4];

3）基于Android手机内置传感器检测法[5]。

本文将Kinect V2用于室内人体检测，利用骨骼跟踪技术和RGB?D相结合的方法可以进行24 h实时检测，可以克服外界光照条件的变化带来的影响，提高跌倒检测的准确度，能够对跌倒行为进行及时处理，保证老年人的安全。