Direction-Independent Human Behavior Recognition Using Distributed Radar Sensor System and Hybrid Neural Network With Dual-View Attention

Radar-based human behavior recognition has been extensively studied. However, the majority of current researches in this area are based on the use of a single radar system. Single radar-based human behavior recognition suffers from radar detection blindness and perspective challenges. Consequently, we present a distributed radar sensor system that addresses the shortcomings of single radar-based human behavior recognition in this article. The proposed distributed radar sensor system adopts six radars and employs an asymmetric layout, which reduces the interaction of individual radars and eliminates the non-detection zone. Moreover, in view of the inhomogeneous distribution of features in the velocity cloud maps between distributed radar views, a recognition network dedicated for distributed radar sensor systems and named the hybrid neural network with dual-view attention (HNN-DVA), has been designed based on the physical meaning of distributed radar velocity point cloud maps. Experimental results show that HNN-DVA has excellent recognition accuracy on our constructed fall-related human behavior datasets.