一种基于室内场景的多摄像头人体定位算法  

Multi-Camera Scene Human Body Location Tracking Algorithm

作　　者：石强 葛源[1] 袁志敏[1] 韩娟娟 崔养浩 Shi Qiang;Ge Yuan;Yuan Zhimin;Han Juanjuan;Cui Yanghao(Nuclear Power Institute of China,Chengdu,610005,China)

机构地区：[1]中国核动力研究设计院,成都610005

出　　处：《仪器仪表用户》2020年第11期24-28,12,共6页Instrumentation

摘　　要：室内场景中的人体定位追踪,通常用于智能监控和室内服务机器人的人机交互。传统多摄像头场景人体定位算法思想来源于三维重建,缺点是需要标定的参数众多,实时性较差。而基于主轴的场景人体定位算法虽然实时性好,但是无法处理人体遮挡的情况。基于像素的人体定位算法虽然精度较高,但是实时性较差。本文重新设计了一种非标定的场景人体定位算法,其算法思想是利用检测得到的人体矩形框代替人体前景信息,在矩形框的上边和下边进行采样单应投影,获取人体候选脚点,进行纬度聚类,在簇内根据到簇中心进行权值授予以及依据权值阈值筛选有效候选脚点,最终获取人体脚点位置。实验证明,本文设计的场景人体定位算法不仅精度高,而且速度快。Human body location tracking in home scenes is commonly used for intelligent monitoring and human-computer interaction of home service robots.The traditional multi-camera scene human body localization algorithm comes from three-dimensional reconstruction.The disadvantage is that there are many parameters to be calibrated,and the real-time performance is poor.The human body localization algorithm based on the spindle is good in real-time,but it cannot handle the occlusion of the human body.Although the pixel-based human body localization algorithm has higher precision,it has poor real-time performance.This paper redesigns an uncalibrated scene human body localization algorithm.The algorithm idea is to replace the human body foreground information with the detected human body rectangular frame.The sampling plane should be projected on the upper and lower sides of the rectangular frame to obtain the human body candidate foot points.Latitude clustering,weighting according to the center of the cluster in the cluster,and screening effective candidate foot points according to the weight threshold,and finally obtaining the position of the human foot.Experiments show that the human body localization algorithm designed in this paper is not only high precision,but also fast.

关 键 词：场景定位 采样 单应投影 权值授予 有效点选取 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]