A human action recognition method based on text supervision
By analyzing joint connections using a skeletal encoder and a graph Transformer module, and combining this with a text encoder to generate feature vectors, the problem of ignoring joint connections in existing technologies is solved, thereby improving the accuracy and generalization ability of human motion recognition.
CN116403278BActive Publication Date: 2026-06-09ZHEJIANG UNIV OF TECH
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- ZHEJIANG UNIV OF TECH
- Filing Date
- 2023-03-22
- Publication Date
- 2026-06-09
Smart Images
Figure CN116403278B_ABST
Abstract
The application relates to the computer technical field and particularly relates to a human action recognition method based on text supervision; first, skeleton data is subjected to human skeleton feature extraction through an encoder network model, then a feature vector of different actions is generated by using a text encoder, similarity calculation is carried out on the feature vector and the skeleton feature, and an action recognition result is obtained; the skeleton feature vector is obtained through the skeleton encoder, a graph Transformer module is added in front of a graph convolution network, the connection between unconnected nodes can be analyzed, compared with an action analysis method based on RGB, the generalization ability is effectively increased, and finally the skeleton feature is supervised and learned through the text feature obtained through the text encoder, so that the action recognition accuracy is effectively improved.
Need to check novelty before this filing date? Find Prior Art