Hand Gesture Recognition Method based on Dual-Channel Residual Neural Network

Abstract

With the changes in the international situation, more stringent requirements have been put forward for counter-terrorism and stability maintenance tasks. Under invisible conditions such as night and shelter, the recognition of gestures becomes very important. The radar has the characteristics of all-weather and has a certain penetration ability. Therefore, radar has an irreplaceable role in solving this problem. Since wideband radar has higher range resolution than narrow-band radar, it can extract finer features of gestures, which has become the main sensing apparatus for gesture recognition. However, gestures are often in a complex background. Furthermore, gestures weak compared to the main body. Therefore, the classification and recognition of human micro-motion based on wideband radar is still a difficult problem. Inspired by the successful application of residual neural network in computer vision, this paper proposes a wideband radar gestures recognition method based on Pyramid Split Attention(PSA) dual-channel residual neural network, which takes the Range-Doppler (RD) map and High Resolution Range Profiles (HRRPs) of human micro-motion as inputs. The effectiveness of this method is verified by experimental data, after the information is convolved, the features are fused, and finally the purpose of classification is achieved. The target recognition rate of this method is 97.83%, which is higher than 95.83% without pyramid split attention, and much higher than 92.33% of the high-resolution range profile based recognition method and 90% of the range Doppler based recognition method, showing the superiority of the proposed method.