计算机工程与应用 ›› 2017, Vol. 53 ›› Issue (20): 173-179.DOI: 10.3778/j.issn.1002-8331.1604-0423

• 图形图像处理 • 上一篇    下一篇

改进混合高斯模型在人体跌倒检测中的应用

孙  朋1,2,夏  飞1,2,3,张  浩1,2,3,彭道刚1,2,马  茜1,2,罗志疆1,2   

  1. 1.上海电力学院 自动化工程学院,上海 200090
    2.上海发电过程智能管控工程技术研究中心,上海 200090
    3.同济大学 CMIS中心,上海 200092
  • 出版日期:2017-10-15 发布日期:2017-10-31

Research of human fall detection algorithm based on improved Gaussian mixture model

SUN Peng1,2, XIA Fei1,2,3, ZHANG Hao1,2,3, PENG Daogang1,2, MA Xi1,2, LUO Zhijiang1,2   

  1. 1.College of Automation Engineering, Shanghai University of Electric Power, Shanghai 200090, China
    2.Shanghai Engineering Research Center of Intelligent Management and Control for Power Process, Shanghai 200090, China
    3.CMIS Center, Tongji University, Shanghai 200092, China
  • Online:2017-10-15 Published:2017-10-31

摘要: 提出了一种基于改进混合高斯模型的分级特征检测算法,对人体跌倒状态进行检测。针对实际目标检测过程中背景更新缓慢,阴影干扰等缺点,通过改进混合高斯模型进行背景更新,并根据阴影区域在HSV颜色空间的特征信息消除阴影干扰。利用人体最小面积外接矩形和垂直外接矩形对检测到的人体目标进行标记,分析目标区域的矩形宽高比、人体质心高度比和人体躯干倾斜角的特征变化。根据各个特征对人体不同状态的判断灵敏度,提出了一种分级特征检测的方法。首先通过人体躯干倾角的特征,判断出人处于非站立状态。接下来依次采用人体质心高度比和矩形宽高比的特征,确认人体处于跌倒状态。实验结果证明,采用提出的方法对人体跌倒状态进行检测,其环境适应性和跌倒检测准确率均高于采用背景差分和直接检测的各种方法。

关键词: 混合高斯模型, HSV, 分级特征, 跌倒检测

Abstract: A hierarchical feature detection based on improved Gaussian mixture model is proposed in this paper to detect the falls. Aiming at the drawbacks of low background updating rate and shadow interference, background model is updated by improving the Gaussian mixture model and the shadow interference is eliminated through the characteristics of shadows in the HSV color space. Human minimum area external rectangle and vertical external rectangle are used to detect the human body, and the characteristics of rectangle ratio, centroid height ratio and inclination angle are analyzed in the paper. A hierarchical feature detection is proposed based on the different sensitivities of the characteristics. Firstly, the non-upright postures are detected by using human inclination angle. Secondly, the falls are confirmed by using the centroid height ratio and rectangle ratio successively. The experimental results show that the fall detection algorithm proposed in this paper has preferable adaptive performance and higher fall detection accuracy compared with different background subtraction algorithms and direct detection algorithms.

Key words: Gaussian mixture model, HSV, hierarchical feature, fall detection