When a human body moves within the coverage range of Wi-Fi signals, the reflected Wi-Fi signals by the various parts of the human body change the propagation path, so analysis of the channel state data can achieve the perception of the human motion. By extracting the Channel State Information (CSI) related to human motion from the Wi-Fi signals and analyzing it with the introduced machine learning classification algorithm, the human motion in the spatial environment can be perceived. On the basis of this theory, this paper proposed an algorithm of human behavior recognition based on CSI wireless sensing to realize deviceless and over-the-air slide turning. This algorithm collects the environmental information containing upward or downward wave in a conference room scene, uses the local outlier factor detection algorithm to segment the actions, and then the time domain features are extracted to train Support Vector Machine (SVM) and eXtreme Gradient Boosting (XGBoost) classification modules. The experimental results show that the average accuracy of the XGBoost module sensing slide flipping can reach 94%, and the SVM module can reach 89%, so the module could be extended to the field of smart classroom and significantly improve speech efficiency.