Driver behavior recognition method and system based on skeleton and semantic information mixing

The invention discloses a driver behavior recognition method and system based on skeleton and semantic information mixing, and relates to the technical field of driver behavior recognition. A behavior recognition method based on semantic information and skeleton information is fused through a space soft attention mechanism based on driver skeleton key points, and driver behaviors are recognized. Firstly, a spatial soft attention module based on driver skeleton key point information is designed, and the module endows the semantic information of the hands and the head of a driver with a large weight to suppress global background information so as to pertinently improve the semantic feature extraction performance related to driver behaviors; designing a hybrid network based on driver skeleton and semantic information, wherein the hybrid network is used for extracting semantic and time features related to driver behavior recognition; in addition, in a night driver behavior recognition task, an adaptive histogram