HIGHLY ROBUST ACTION RECOGNITION METHOD BASED ON CHANNEL STATE INFORMATION

The methods based on Wi-Fi technology are increasingly popular research directions in the field of action recognition due to its advantages of unwearable and easy deployment. However, in the case of interference, Wi-Fi devices are vulnerable to influence, resulting in a decrease in identification accuracy. Accordingly, a highly robust action recognition method based on channel status information (CSI) is designed. The dynamic subcarrier selection algorithm was proposed to dynamically select the most relevant subcarrier. In view of the problem of poor data acquisition quality and inaccurate segmentation of the action recognition accuracy under interference conditions, the segmentation auxiliary algorithm was proposed to effectively improve the segmentation accuracy and classification accuracy of the action interval. Experimental results show that this method achieves recognition accuracy of 92% and 81% for five types of actions in both interference and interference free environments, respectively, demonstrating strong robustness.