人体背部动态识别与跟踪方法

Dynamic recognizing and tracing for the back surface of the human body

为了准确、快速地跟踪人体,本文以人体背部为研究对象,提出了一种融合空间运动变换矩阵及点云粗-精配准算法的动态识别与跟踪方法。首先,采用直通滤波和统计滤波以及背景去除分割初始场景点云,识别出目标人体,再基于人体背部几何特征,利用微元分割法进行背部划分(即获得目标区域);其次,提取内部形状描述子(ISS)关键点简化背部点云,并结合快速点特征直方图(FPFH)通过采样一致性(SAC-IA)算法和迭代最接近点(ICP)算法进行点云配准,并通过相邻两帧点云配准结果,获得背部空间运动变换矩阵。通过分别与3Dcs-ICP算法和普通粗-精配准算法跟踪实验发现,本文提出算法运行时间明显减少,目标区域在X、Y和Z方向的平均均方根误差分别为0.264 cm、0.261 cm和0.517 cm。实验结果表明:此方法可提高人体背部识别速度和准确度,也为其他人体器官的跟踪与识别提供参考。

This paper proposes a dynamic identification and tracking method integrating a spatial motion transformation matrix and point cloud coarse-accurate alignment algorithm with the back as the research object,In order to accurately and quickly trace the human body.Straight-pass filtering,statistical filtering and background removal are used to segment the initial field attraction cloud to identify the target human body,and then based on the geometric features of the back,the back is divided using the micro-element segmentation method(i.e.,obtain target area).The Intrinsic Shape Signature(ISS)key points are extracted to simplify the dorsal point cloud and combined with the Fast Point Feature Histogram(FPFH)to align the point cloud by the Sample Consensus Initial Alignment(SAC-IA)algorithm and Iterative Closest Point(ICP)algorithm.The backspace motion transformation matrix is obtained from the alignment results of two adjacent frames of the point cloud.The runtime results show that the proposed algorithm can reduce the runtime significantly by tracking with the 3Dcs-ICP algorithm and standard coarse-fine alignment algorithm.The average root mean square error of the target area in the X,Y and Z directions are 0.264 cm,0.261 cm and 0.517 cm,respectively.The experimental results show that this method can improve the speed and accuracy of the back identification and provide a reference for tracking and identifying other human organs.