This study proposes a method specifically designed for recognising students' dance teaching movements based on spatial temporal graph convolutional networks.The method employs three parallel spatial temporal aggregation graph convolutions to extract dynamic features.By incorporating OpenPose for human pose estimation, the proposed spatial temporal graph convolutional network-based action recognition method effectively classifies student dance teaching movements.Experimental model achieving recognition accuracies of 89.1% and 90.4% on cross-subject and cross-set benchmarks respectively, compared with existing advanced models, the improvement ranges from 0.3 to 4.0 percentage points and 0.3 to 3.6 percentage points respectively.Furthermore, evaluation of the self-constructed dataset encompassing ten classical dance categories demonstrates that the method achieves recognition accuracy ranging from 89.6% to 96.1%, effectively resolving the issue of inaccurate dance movement recognition.
Lin Ma (Thu,) studied this question.