| Title |
Dance Movement Recognition based on Deep Learning |
| DOI |
https://doi.org/10.5573/IEIESPC.2024.13.3.209 |
| Keywords |
Deep learning; Convolutional neural network; Movement recognition |
| Abstract |
In recent years with continuous development of the computer vision field, there has been an increasing demand for fast and accurate recognition of human movement, especially in sports. This paper researches ballet movements, which are recognized and analyzed using a convolutional neural network (CNN) based on deep learning. Training of the CNN is improved by particle swarm optimization (PSO). Then, 1,000 ballet videos are used as a dataset to compare optimized CNN, traditional CNN, and support vector machine (SVM) methods. The results show that the improved CNN converged fastest, stabilizing after about five iterations, whereas the traditional CNN method took approximately 20 iterations to stabilize. Additionally, after convergence, error in the improved CNN was smaller than from the traditional CNN. The average recognition accuracy of the SVM method was 84.17%, with a recognition time of 3.32 seconds; for the traditional CNN method, it was 90.16% with a recognition time of 2.68 s; and for the improved CNN method, it was 95.66% with a recognition time of only 1.35 s. |