Action Evaluation Based on Monocular RGB Video

View Paper (Chinese)
  • Introduced new metrics for video pose evaluation and realized action counting for gym actions.
  • Deployed OpenPose and ST-GCN for human pose estimation and action recognition.
  • Designed an action segmentation method based on keyframe extraction and clustering (PCA & K-means).
  • Proposed a weak-rigid transformation for pose registration and defined a pose feature descriptor.