Deep reinforcement learning-supported detection and correction system for young children's dance movements

Motor skill development through dance plays an important role in improving children’s coordination, rhythm, balance, and body awareness. However, young children often struggle to perform dance movements accurately because they receive limited personalized and immediate feedback during practice. To address this challenge, this study proposes an intelligent dance movement detection and correction system that automatically identifies movement errors and provides real-time corrective guidance. The proposed framework combines multimodal motion analysis with a Deep Reinforcement Learning (DRL) strategy to continuously evaluate movement quality and optimize corrective feedback. The proposed framework integrates an Adaptive Runge–Kutta optimized Twin Delayed Deep Deterministic Policy Gradient (AdapRK-TD3PG) algorithm with multimodal feature representation, enabling stable learning. A Dance Movement Learning Dataset comprising 10,000 samples from children aged 5–12 years, which includes synchronized video and audio data, along with extracted skeletal and kinematic features. Preprocessing is performed using Gaussian noise reduction to smooth joint trajectories and temporal normalization to standardize sequence length and feature distribution, ensuring consistent spatial–temporal representation. Feature extraction is carried out using a lightweight Convolutional Neural Network (CNN) to capture keypoints, joint positions, and limb trajectories. These features are further enhanced with joint angle information from OpenPose to model both spatial and temporal dynamics of movement. Experimental setup was developed by using python and results demonstrate reduced synchronization error (MSE: 1.67), Multilabel Dance Classification of 86.73, and aesthetic accuracy of 97.56%. The model achieves low prediction errors (MSE: 0.009, MAE: 0.013), and an inference time of 4.3 ms. Overall, proposed approach enables effective motor skill development.