Dance Learner Posture Estimation Using an Improved Version of the Higher Hrnet Network

Abstract

Dance posture estimation helps train better results and reduces risks while making dance movements more effective. Regular instructor control lack’s reliability because their judgments vary. Our study develops an updated HRNet system for dance posture estimation with features focused on dance movement understanding and dynamic pattern learning. By comparing the proposed model with OpenPose and PoseResNet the research demonstrates better accuracy, more reliable poses, and real-time results in training outcomes. The tests produced 94.6% mAP while decreasing angular errors to 4.9 degrees and keeping tracking stable at 96.8%. This setup fulfills AI-based dance training needs effectively.

Keywords: Dance posture estimation, Deep Learning, Dance training, Real- time analysis, Computer Vision, HRNet.