欢迎访问《图学学报》 分享到：
图学学报

图学学报 ›› 2022, Vol. 43 ›› Issue (5): 858-864.DOI: 10.11996/JG.j.2095-302X.2022050858

• 图像处理与计算机视觉 • 上一篇    下一篇

基于视频的人体行走参数测量方法及应用

  

  1. 1. 东南大学仪器科学与工程学院，江苏 南京 210096；  2. 杭州萤石软件有限公司，浙江 杭州 310051
  • 出版日期:2022-10-31 发布日期:2022-10-28
  • 基金资助:
    国家重点研发计划项目(2019YFC511501)

Video-based human walking parameter measurement method and application

  1. 1. School of Instrument Science and Engineering, Southeast University, Nanjing Jiangsu 210096, China;  2. Hangzhou EZVIZ Software Co., Ltd., Hangzhou Zhejiang 310051, China
  • Online:2022-10-31 Published:2022-10-28
  • Supported by:
    National Key Research and Development Program (2019YFC511501)

PDF (PC)

54

摘要：

针对现有的人体行走参数测量方法复杂度高、效率低等问题，提出了一种基于视频的人体行走 参数测量方法。利用监督学习的方法对视频中的运动目标进行姿态估计，逐帧识别骨骼关节点。然后根据头部 和脚部特征点，结合场景标定获取的像素距离与实际距离的转换关系，实现行走身高测量；根据关节特征点， 利用余弦公式计算关节活动度；根据脚部特征点，提出了一种结合前后极点帧差和像素差判断行走步长和步速 的方法。最后提出了一种基于 Unity3D 的虚拟人随动控制方法，能够在虚拟场景中进行运动仿真，便于实时监 控和分析视频中的人体异常行为并做出预警。实验表明该方法具有操作简单、准确度高和实时性强等优点。

关键词: 单目视频, 人体关节点, 行走参数, 虚拟人仿真, 异常行为

Abstract:

To address the problems of high complexity and low efficiency of the available methods for human walking parameter measurement, a video-based human walking parameter measurement method was proposed. The supervised learning method was used to estimate the pose of the moving target in the video, and the bone joint points were recognized frame by frame. Then, according to the feature points of head and feet, combined with the conversion relationship between the pixel distance obtained by scene calibration and the actual distance, the walking height measurement was realized. According to the joint feature points, the joint range of motion was calculated by cosine formula. According to the foot feature points, a method to identify the walking step size and pace was proposed by combining the frame difference of front and rear poles and the pixel difference. Finally, a virtual human follow-up control method based on Unity3D was proposed, which could carry out motion simulation in virtual scenes and was convenient for real-time monitoring and human abnormal behavior analysis in videos and early warning issuing. Experiments show that this method is superior in simple operation, high accuracy, and strong real-time performance. 

Key words: monocular video, human joint points, walking parameter, virtual human simulation, abnormal behavior

中图分类号: