Wi-Do: Highly Robust Human Motion Perception Model Under WiFi Signal

Human-computer interaction is an important way for the Internet of Things to become intelligent, and human motion recognition has become the core technology for the realization of intelligent environments. Due to its good user experience, high universal performance, and low deployment cost, WiFi-based human motion recognition technology stands out from many interactive technologies and has shown great performance in the fields of smart security, sports health, and elderly active detection. In the existing WiFi motion recognition work, the motion recognition feature changes due to the change of the direction. In order to ensure the recognition accuracy, the motion recognition direction needs to be fixed. This directional dependence has caused a great obstacle to the wider interactive application of WiFi-based motion recognition technology. To overcome this limitation, a direction-independent motion recognition solution, Wi-Do, is proposed. It is a highly robust human motion perception model under WiFi signals. In this work, the antenna diversity is used to eliminate the random phase shift; the Doppler shift and the fast Fourier transform value are used as the recognition features; and the bidirectional GRU of attention mechanism is introduced to classify and recognize the motion. This model integrates spatial features into the time model, which improves the robustness and accuracy of wireless signal recognition of human actions. The results of the experiment in a typical indoor environment show superior performance and 93% accuracy, verifying that the method is significantly better than the previous method.