Human Activity Recognition Based on the Method of CFAR-SOPC Using Millimeter-Wave Radar

Human activity recognition (HAR) has become a research hotspot due to its broad application prospects in search and rescue, health monitoring, safety monitoring, and sports science. Millimeter-wave radar, with its low cost and noncontact sensing method, provides an ideal technical solution for protecting user privacy, so it is widely used in HAR research. In the study, a human motion classification method based on the constant false alarm rate-sampling the overall point cloud (CFAR-SOPC) is proposed. First, the human motion data from a millimeter-wave radar are resized into a 2-D cube. Second, the phasor average cancellation (PAC) method is applied to the data to filter out clutter, and then, the data are resized into a 3-D cube, and CFAR-SOPC is performed on the data to generate a range–Doppler (RD)–time stereo contour point cloud (SCPC), which effectively reduces the size of the features. Finally, the sample is input into the PointNet network that specializes in processing point cloud data for feature extraction and activity recognition, with an accuracy rate of 96.7%. The experimental results present a fact that compared with the existing approaches, CFAR-SOPC improves the accuracy of classification and reduces the cost of memory and time.