《计算机应用》唯一官方网站 ›› 2022, Vol. 42 ›› Issue (1): 294-301.DOI: 10.11772/j.issn.1001-9081.2021020331

• 前沿与综合应用 • 上一篇    

基于AlphaPose优化模型的老人跌倒行为检测算法

马敬奇1, 雷欢1, 陈敏翼2()   

  1. 1.广东省现代控制技术重点实验室(广东省科学院智能制造研究所),广州 510070
    2.广东省科学技术情报研究所,广州 510070
  • 收稿日期:2021-03-05 修回日期:2021-05-17 接受日期:2021-05-18 发布日期:2021-06-11 出版日期:2022-01-10
  • 通讯作者: 陈敏翼
  • 作者简介:马敬奇(1988—),男,河南安阳人,工程师,硕士,主要研究方向:行为分析、深度学习
    雷欢(1987—),男,湖南衡阳人,工程师,硕士,主要研究方向:深度学习、图像识别
    陈敏翼(1986—),男,广东潮州人,助理研究员,硕士,主要研究方向:深度学习、大数据分析。
  • 基金资助:
    广州市科技计划项目(202007040007)

Fall behavior detection algorithm for the elderly based on AlphaPose optimization model

Jingqi MA1, Huan LEI1, Minyi CHEN2()   

  1. 1.Guangdong Key Laboratory of Modern Control Technology (Institute of Intelligent Manufacturing,Guangdong Academy of Sciences),Guangzhou Guangdong 510070,China
    2.Guangdong Institute of Scientific and Technical Information,Guangzhou Guangdong 510070,China
  • Received:2021-03-05 Revised:2021-05-17 Accepted:2021-05-18 Online:2021-06-11 Published:2022-01-10
  • Contact: Minyi CHEN
  • About author:MA Jingqi, born in 1988, M. S., engineer. His research interests include behavior analysis, deep learning.
    LEI Huan, born in 1987, M. S., engineer. His research interests include deep learning, image recognition.
    CHEN Minyi, born in 1986, M. S., research assistant. His research interests include deep learning, big data analysis.
  • Supported by:
    Guangzhou Science and Technology Program(202007040007)

摘要:

针对在低功耗、低成本硬件平台快速准确检测老人跌倒高危行为的问题,提出了一种基于AlphaPose优化模型的老人异常行为检测算法。首先,对行人目标检测模型和姿态估计模型进行优化,以加快人体目标检测和姿态关节点推理;然后,通过优化的AlphaPose模型快速计算得到人体姿态关节点图像坐标数据;最后,计算人体跌倒瞬间头部关节点线速度与胯部关节线速度之间的关系,以及人体中垂线与图像X轴之间夹角的变化来判断跌倒现象的发生。将所提算法移植到Jetson Nano嵌入式开发板上,并与当前主要的基于人体姿态的跌倒检测算法YOLOv3+Pose、YOLOv4+Pose、YOLOv5+Pose、trt_pose和NanoDet+Pose进行对比分析。实验结果表明,在所用嵌入式平台上,图像分辨率为320×240时,所提算法的检测帧率达到8.83 frame/s,准确率为0.913,均优于对比算法。该算法具有较高的实时性和准确率,能够及时检测老人跌倒行为的发生。

关键词: 实时跌倒检测, 姿态估计, 姿态关节点, 嵌入式平台, 目标检测, 深度学习

Abstract:

In order to detect the elderly fall high-risk behaviors quickly and accurately on the low-power and low-cost hardware platform, an abnormal behavior detection algorithm based on AlphaPose optimization model was proposed. Firstly, the pedestrian target detection model and pose estimation model were optimized to accelerate the human target detection and pose joint point reasoning. Then, the image coordinate data of human pose joint points were computed rapidly through the optimized AlphaPose model. Finally, the relationship between the head joint point linear velocity and the crotch joint linear velocity at the moment the human body falls was calculated, as well as the change of the angle between the midperpendicular of the torso and X-axis of the image, were calculated to determine the occurrence of the fall. The proposed algorithm was deployed to the Jetson Nano embedded development board, and compared with several main fall detection algorithms based on human pose at present: YOLO (You Only Look Once)v3+Pose, YOLOv4+Pose, YOLOv5+Pose, trt_pose and NanoDet+Pose. Experimental results show that on the used embedded platform when the image resolution is 320×240, the proposed algorithm has the detection frame rate of 8.83 frame/s and the accuracy of 0.913, which are both better than those of the algorithms compared above. The proposed algorithm has relatively high real-time performance and accuracy, and can timely detect the occurrence of the elderly fall behaviors.

Key words: real-time fall detection, pose estimation, pose joint point, embedded platform, target detection, deep learning

中图分类号: