Graph convolutional neural network action recognition method based on dynamic time warping, electronic equipment and storage medium

The invention discloses a graph convolutional neural network action recognition method based on dynamic time warping, electronic equipment and a storage medium, and the method comprises the steps: constructing an RGB video into graph structure data, and calculating the importance of the spatial feature of each node in a graph data structure; the time dynamic change is quantified; remapping the time axis coordinates; generating a new remapping feature based on time axis dynamic scaling; constructing a human-centered action time convolutional network and an object-centered affordability time convolutional network; performing action classification and object availability classification; and obtaining the final action classification score of the person and the availability score of the object in each candidate box. According to the graph convolutional neural network action recognition method based on dynamic time warping, a video feature warping method in the time axis direction is proposed for the first time, in