A Real-time Human Activity Recognition Method for Through-the-Wall Radar

Human activity recognition (HAR) has long been a research hotspot in anti-terrorism, patient monitoring and other applications. Throughout the current research progress, the realtime recognition of unknown start and end time has not been well solved. In addition, the recognition accuracy of blocking human in the wall-through scene needs to be improved. To tackle this issue, this paper proposed a novel range profile sequence driven end-to-end model, which specifically employed random cropping training method. Then, we carried out experiments on actual Through-the-Wall Radar(TWR), finally achieved an average accuracy of 97.6% on four common activities, and the most significant is that our method can immediately output recognition results, without waiting for the end of activity.