Research On Track And Fall Detection Algorithm Of Indoor Targets By Millimeter Wave Radar

With the increase in life expectancy and the decline in birth rates,the problem of population aging is becoming more and more serious.The injury caused by falls is a serious threat to the life safety of the elderly,so the elderly fall detection has attracted social attention.In this paper,the fall detection technology of indoor target tracking is studied based on millimeter wave radar.This paper mainly includes the following work:1)An indoor target tracking and fall detection system based on millimeter wave radar is established.After testing,the installation height of millimeter wave radar applicable to this paper is determined to be 1.5-2.5m,with a downward tilt of 10-15 degrees,and the configuration parameters with the best range resolution are selected.Millimeter wave radar signal data is processed by Fourier transform and CFAR in the integrated processor to obtain point cloud data.The host computer software is designed to collect and display point cloud data in real time as the data input of point cloud clustering algorithm.The true and false target clustering clusters are obtained by point cloud clustering algorithm.The target tracking algorithm filters out the false clustering clusters and predicts and tracks the trajectories of the true clustering clusters.Through threshold classifier,fall detection is carried out on the target which has formed tracking,and the result is displayed on the interface of upper computer software.2)An improved point cloud clustering algorithm based on DBSCAN is proposed.The SNR data of point cloud is added into the input of DBSCAN to improve the similarity between points in point cloud cluster.By analyzing the point cloud data,it is determined that the point cloud shape is approximately spherical when the target moves,so as to extend the two-dimensional space of the original algorithm to threedimensional space and improve the accuracy of point cloud clustering.The result of the improved clustering algorithm deviates only about 0.141 m from the actual XY position of the target,which is more accurate than the original DBSCAN result.3)Select the applicable target tracking algorithm.The simulation results show that the improved Hough transform method has better simulation effect and strong stability,the extended Kalman filter algorithm has less error fluctuation and higher computational efficiency,and the nearest neighbor data association algorithm has higher computational efficiency.In this paper,the improved Hough transform,extended Kalman filter and nearest neighbor data association are used to track the target.4)A fall detection method based on threshold classification is proposed.The average value of 5 frames of target tracking data was selected as the action key points to analyze the point cloud data of target action key points and extract X,Y,Z and V features to form feature vectors.Considering the fluctuation of threshold parameters and analyzing threshold parameters through testing,a threshold based fall detection is constructed.The actual test results show that the accuracy of falling,sitting and standing motions is 88.6%,84.5% and 94.6%,respectively.