A Hybrid Deep Learning Model for Dance Recognition with Applications in Robotic Vision and AI Automation

Abstract

Image recognition has gained significant attention due to advancements in artificial intelligence (AI) and machine learning. Traditional dance instruction often relies on subjective feedback, leading to inconsistencies in assessment. This paper presents a hybrid deep learning model integrating Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks for dance movement recognition. CNNs extract spatial features, while LSTMs capture temporal dependencies, enabling accurate recognition of dance sequences. The dataset used combines ballet, locking, and popping dance genres, with preprocessing techniques such as augmentation and normalization. The hybrid model achieves 99.52% accuracy, surpassing standalone CNN and LSTM models. Evaluation metrics, including Matthews Correlation Coefficient (MCC) and AUC-ROC, validate the model's ability to distinguish complex movement patterns with high reliability. Beyond dance recognition, precise motion classification plays a crucial role in robotic vision and AI-driven automation, where accurate pose estimation and movement tracking enable applications in humanoid robot motion imitation, gesture-based human-robot interaction, and precision automation in industrial workflows. Future work will focus on real-time deployment, dataset expansion, and integration with augmented reality to enhance both dance training and humanoid robot motion learning.