Combined action recognition method and system

The invention provides a combined action recognition method and system, and belongs to the technical field of computer vision, and the method comprises the steps: obtaining the geometric position information of an object contained in a to-be-recognized video; obtaining multi-modal features of each object based on the geometric position coordinates; wherein the multi-modal features comprise geometric characterization features, visual characterization features and motion flow features; according to the multi-modal features, obtaining feature vectors representing interaction relationships among the objects; and based on the feature vectors representing the interaction relationship between the objects, performing action classification to obtain a final combined action recognition result. According to the method, through a Transform module based on a self-attention mechanism, the interaction relation of different objects in a video sequence in a time domain and a space domain is deduced, RGB visual representation