Continuous sign language recognition method based on I3D-TCP network structure

The invention discloses a continuous sign language recognition method based on an I3D-TCP network structure, and the method comprises the steps: carrying out the sign language information feature extraction of an RGB sign language video through employing an I3D model, capturing the dynamic information of the sign language video through employing a time attention module, carrying out the conversion of a sign language video time sequence into a covariance pool matrix, the correlation of the content of a frame, and the correlation between frames, and carrying out the recognition of the sign language information. And finally, network training is accelerated by using matrix power normalization, the generalization ability of the network is improved, and sign language recognition in a complex scene is better realized. The method is seamlessly connected with an existing I3D model, an effective sign language video recognition framework is generated, sign languages can be accurately recognized, the generalization abili