Behavior recognition method driven by time-sharing and cross-domain asynchronous fusion

The invention discloses a behavior recognition method driven by co-time and cross-domain asynchronous fusion. The behavior recognition method comprises the following steps: acquiring multi-person body color images as an image sequence; estimating a two-dimensional human body posture from the intercepted image, and calculating a key point heat map of the two-dimensional human body posture as a posture sequence; establishing an apparent network, taking the image sequence as input, and extracting image features; meanwhile, establishing an attitude network, taking the attitude sequence as input, and extracting attitude features; taking the image features and the attitude features as input, and zooming, aligning and fusing the image features and the attitude features; normalizing the fused image features and attitude features, splicing the normalized double-flow features to obtain global features, and inputting the global features into a classifier to solve a behavior category with the maximum probability; a multi