Wi-Fi Signal Gesture Recognition Based on Multimodality

With the development of technology, various gesture recognition devices and technologies have emerged to meet people's various needs. Traditional gesture recognition methods are relatively cumbersome and require wearing data gloves. At the same time, the technology based on computer vision needs to recognize that the target is always within sight. Therefore, this paper proposes a gesture recognition method based on multimodal Wi-Fi signals, and optimizes the gesture recognition methods from devices and technologies to overcome the problems in traditional recognition methods. Specifically, we first built a Wi-Fi signal data acquisition platform based on the Atheros network card, and the packet loss rate is less than 0.1%. Then, frame extraction is carried out for the video signal, and the T3D network based on DenseNet is used for video recognition. The video recognition rate reaches 95.0%. Finally, the above video data is extended to the Wi-Fi signal, and gesture recognition is performed jointly with CNN and GRU. The test data shows that the gesture recognition rate of the system platform is 88.2%, which is the leading level.