基于毫米波雷达的室内老人轨迹追踪算法  

Trajectory tracking algorithm based on millimeter wave radar for indoor elderly man

作　　者：刘树博[1] 赖招宇 罗先喜[1] 李跃忠[1] LIU Shubo;LAI Zhaoyu;LUO Xianxi;LI Yuezhong(Jiangxi Industrial Technology Research Institute of Rehabilitation Assistance,School of Mechanical and Electronic Engineering,East China University of Technology,Nanchang 330013,China)

机构地区：[1]东华理工大学、机械与电子工程学院、江西省康复辅具产业技术研究院,江西南昌330013

出　　处：《实验技术与管理》2023年第7期70-78,共9页Experimental Technology and Management

基　　金：国家自然科学基金项目(62141102);江西省自然科学基金项目(20202BAB202008);教育部第二批协同育人项目(202102405010);江西省重点研发计划项目(20212BBE53033)。

摘　　要：为了实现对室内老人运动状态的实时与准确追踪,提出一种基于毫米波雷达的多目标人物轨迹追踪(multi-objective human trajectory tracking,MOHTT)算法。首先利用静态杂波滤除算法移除雷达数据中的环境干扰;其次根据自适应聚类(self-adaptive clustering,SAC)算法实现人物质心位置识别,基于质心在二维平面中的坐标信息确定人物状态向量,利用多目标离散追踪(multi-objective discrete tracking,MODT)算法实现人物运动状态估计;再利用CNN-GRU校正(CNN-GRU correction,CGC)算法对人物质心状态进行实时校正,得出准确的人体质心位置坐标;在此基础上,通过足迹获取(footprint acquisition,FA)算法得出足迹坐标,经拟合操作后给出完整的人物“质心+足迹”运动轨迹。实验结果表明,MOHTT算法对1、2、3、4个人的检测成功率分别为99.8%、95.6%、93.7%和87.3%,对单人位置检测误差的RMSE为0.152m,且在不同类型干扰下仍能提供具有信服力的检测结果,具有较好的实时性、抗干扰性和准确性,表现出良好的轨迹追踪效果。A multi-objective human trajectory Tracking(MOHTT)algorithm based on millimeter wave radar is proposed in order to realize the real-time and accurate state tracking of the indoor elderly.Firstly,static clutter filtering algorithm is used to delete the environmental noises existed in the radar data.Secondly,self-adaptive clustering(SAC)algorithm is employed to achieve the recognition of human centroid position,and the human state vector is determined based on the coordinate information of centroid in the two-dimensional plane,and a multi-objective discrete tracking(MODT)algorithm is then developed to estimate human motion state.Finally,CNN-GRU correction(CGC)algorithm is used for real-time correction of the centroid state to obtain the accurate human centroid position coordinates,based on which footprint acquisition(FA)algorithm is employed to obtain the footprint coordinates,and then the complete human motion trajectory of“centroid+footprint”is presented by curve fitting.The experimental results show that the success rates of MOHTT algorithm for detecting 1,2,3,and 4 individuals are 99.8%,95.6%,93.7%,and 87.3%respectively,and the RMSE for single person position detection error is 0.152 m,which can still provide convincing detection results under different types of interference,demonstrating the good real-time performance,anti-interference ability,accuracy and good trajectory tracking effect.

关 键 词：毫米波雷达 自适应聚类 多目标离散追踪 CNN-GRU校正算法 足迹获取算法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]