Multi-sensor data fusion bully behavior detection algorithm

The invention relates to the field of behavior recognition and classification, in particular to a multi-sensor data fusion bullying behavior detection algorithm. The method is mainly used for detecting and identifying whether the user encounters the bullying behavior or not, the defects that in an existing detection mode, camera-based human skeleton detection is limited in coverage range and large in calculation amount are overcome, and the defects that the practicability and the real-time performance are poor when the user actively dials the hotline are overcome. According to the algorithm, a three-layer data filtering recognition method is used, the first layer uses an ID3 decision-making tree to judge the bullying behavior, the second layer uses a data fluctuation analysis mode to judge, and the third layer uses decision-making level fusion of a big data-based CNN convolutional neural network model and a user audio emotion analysis method. And the accuracy rate is preset for each layer. According to the in