A novel approach for enhancing personalized motion training and performance evaluation with Attn-MVAGCN

With the continuous development of intelligent sports training, high-precision posture estimation technology plays a vital role in personalized guidance and sports effect evaluation. However, existing posture estimation methods usually have difficulty in balancing high precision and computational efficiency in complex sports scenes. To this end, this paper proposes a new posture estimation model Attn-MVAGCN, which combines the MobileViT module for image feature extraction, the Adaptive-GCN module for joint relationship modeling, and introduces the attention mechanism to enhance the capture of key posture information. Through experiments on multiple standard datasets (COCO-WholeBody, MPII Human Pose, and NTU RGB+D), the results show that Attn-MVAGCN reduces MPJPE (joint position error) by 10.4% (35.2 mm) and improves AP (average precision) by 6.7% (72.5%), while the inference speed reaches 70.5 FPS, which is significantly better than the existing baseline model. Personalized training experiments further verify the effectiveness of the model in dynamically optimizing athlete posture estimation and improving motion standardization. The advantages of the Attn-MVAGCN model in computational efficiency and real-time performance enable it to run efficiently on mobile devices and smart wearable devices. It has the potential for application on resource-constrained devices and has broad application value in fields such as sports rehabilitation and smart fitness.