Udarata Dance Pose Detection and Validation Using Computer Vision and Deep Neural Networks: A Comparative Study of Static and Dynamic Approaches

Udarata dance, a traditional Sri Lankan art form, requires precise body movements and poses. This research introduces a novel approach for detecting and validating these poses using computer vision and deep neural networks. We explore two methodologies: static pose detection, which compares images against a pre-labeled dataset, and dynamic pose detection, which employs Mediapipe to track body landmarks in videos and an LSTM model to classify sequences of poses as either correct or incorrect. Our findings highlight the effectiveness of these methods in accurately identifying proper Udarata dance poses, offering a valuable tool for dance instruction and cultural preservation. For our experiments, we have defined key angles and heights of body parts and compiled a new Udarata Dance Dataset. The experimental results demonstrate that our algorithms achieve superior performance in pose detection and validation.