Human Behavior Recognition Based on Weighted Voting - Dempster Shafer Algorithm from Fusion of Multiple Radars

Human behaviour recognition technology can be applied to a variety of scenarios such as home elderly monitoring, medical care and accident rescue. Millimeter wave (mmWave) radar can stably detect human behavior while protecting privacy, and its noninvasive nature, high resolution and multi-function of mmWave radar bring new possibilities for behaviour recognition. Compared with single radar, the fusion of multi radar data can increase the detection area and further improve behavior recognition accuracy. In this study, we divided each action into eight datasets based on angle, and used support vector machines to learn micro-Doppler features for each data set. Multiple radar data are then combined for human behavior recognition using weighted voting - Dempster Shafer theory (WV-DST). Furthermore, experiments show that the average motion recognition accuracy of this method is 98.6%, especially when the human body is not directly facing the radar, the recognition accuracy is significantly improved.