基于CNN的WiFi信号复杂动态手势自动检测与识别

WiFi signal has been proven it can be used for dynamic hand gesture recognition, which is expected to provide a novel way for human-computer interaction (HCI). Traditional methods for dynamic hand gesture recognition with WiFi can just recognize simple gestures like up and down movement, left and right movement. Besides, the detection and segmentation algorithms for the gestures are performed in offline data. In this paper, a complete dynamic hand gesture automatic detection and recognition framework is proposed by using the instantaneous received signal strength (IRSS) extracted from multiple independent WiFi nodes. We analyzed that the starting and ending points of the gesture waveform segments are not absolutely accurate no matter by manual segmentation or the proposed auto-segmentation algorithm. Therefore, we designed a recognition module based on convolutional neural network (CNN) to effectively eliminate the errors caused by segmentation phase to improve the recognition accuracy of the system. In the experimental phase, we can adjust the parameters to balance gesture detection accuracy and recognition accuracy. The system can achieve 88.98% recognition accuracy with 87.88% detection accuracy or 94.79% recognition accuracy with 72.12% detection accuracy.