A New Method of Human Gesture Recognition Using Wi-Fi Signals Based on XGBoost

Human gesture recognition has drawn widespread attention for its great application value in both the Internet of Things (IoT) and Human-Computer Interaction (HCI). Although most of the existing approaches have achieved promising effect, they rely on deep learning method enabled by a large number of samples. In this paper, a gesture recognition method based on the eXtreme Gradient Boosting (XGBoost) classification model is proposed to achieve gesture identification without too many samples and features. Meanwhile, it can maintain the recognition accuracy as well as the recognition speed. We collected six predefined dynamic gestures samples and conducted extensive experiments to evaluate its performance. The results demonstrate that our method can achieve an average recognition accuracy of 94.55% when ten features are used and average accuracy of 91.75% when two suitable features are selected. Comparing with the traditional classification algorithms, the method presented in this paper has a great balance among performance, recognition speed, and the number of features of the gestures.