基于空间聚类的FMCW雷达双人行为识别方法

Two-persons activity recognition method for FMCW radar based on spatial information clustering

为实现双人场景下人体行为的识别,利用调频连续波(frequency modulated continuous wave,FMCW)雷达提出一种基于空间聚类的双人行为识别方法.该方法采用基于密度的DBSCAN(density-based spatial clustering of applications with noise)聚类算法将FMCW雷达采集到的坐标数据聚类成不同的聚类群,使得每一个聚类群对应于单一人体的行为,再对其进行数据处理、特征提取后分别采用机器学习方法分类,实现双人场景下人体行为的识别.文中分析行为特征量、动作关键点以及分类器对识别准确率的影响.实验结果表明,在两人场景中该方法对跌倒、坐下和行走的检测准确率分别可以达到100%、93.8%和87.3%.

In order to recognize human behavior in the two-persons scenario,a method based on spatial clustering for frequency modulated continuous wave(FMCW)radar is proposed.In this method,DBSCAN(density-based spatial clustering of applications with noise)clustering algorithm is used to cluster the coordinate data collected by FMCW radar into different clustering groups,so that each clustering group corresponds to the behavior initiated by a single human body,and then machine learning method is used to classify them respectively after processing data,extracting features and so as to realize identification of body behavior under two-persons scenario.This paper analyzed the influence of behavior features,key points and classifiers on recognition accuracy.The experiments show that the recognition accuracy of the three activities:fall,sitting and walking in 2-person scenario can reach 100%,93.8%and 87.3%respctively.