Traditional turkish folk dance figure detection using deep learning: The case of Harmandalı

Folk dances are a vibrant expression of a culture's spirit and identity, carrying the weight of history, traditions, and values. This thesis explores the use of computer vision and deep learning to enhance the teaching and evaluation of the Turkish Folk Dance Harmandalı, bridging tradition with modern technology to preserve this rich cultural heritage. A sophisticated system was developed to analyze dance videos of both students and teachers. Utilizing Mediapipe and YOLO human pose estimation algorithms, skeletal data is extracted and processed through Affine transformation. This data is then rigorously compared using a variety of methods including time-lagged cross-correlation, long short-term memory networks (LSTM), Euclidean distance, cosine similarity, and dynamic time warping (DTW). The result is an objective assessment of student performance, offering valuable insights into areas for improvement. The system's user-friendly interface provides students with visual feedback and a tailored learning experience. Teachers benefit from detailed performance analysis, aiding in personalized instruction. Research findings demonstrate the superiority of the LSTM method in aligning with expert evaluations, achieving an average score of 68.43 and an MSE of 56.11, closest to the expert average of 66.67. DTW followed as the second most successful model. This thesis underscores the importance of technology integration in traditional dance education, while also offering a new perspective on the analysis and evaluation of Turkish Folk Dances, with the aim of contributing to the preservation of cultural heritage.