Automatic Transcription of Greek Folk Dance Videos to Labanotation Based on Autoencoders

We’re creating an automatic system, based on autoencoders, to transcribe dance videos into labanotation [1], a movement notation system. Manual labanotation generation is a time-consuming process that requires specialized knowledge. Our system aims to save time and provide a valuable tool for choreographers, dancers, and anyone interested in documenting body movement. Our system analyzes RGB videos of dancers, isolates their movements, and generates labanotation as an image. The process involves extracting the 3D skeleton, segmenting movements, identifying them, and mapping them to Laban symbols. In our research, we focus on segmenting the movements of the dancer’s lower body. We calculate the angles of the legs and use them as features to train an autoencoder. This approach, inspired by [2], has not been previously explored for human movement segmentation. Human movement segmentation remains a hard problem due to the temporal complexity among the high-dimensional motion features. Our work aims to automatically generate labanotation for Greek folk dances, contributing to the preservation and transmission of dance-related Intangible Cultural Heritage (ICH). The system is integrated into the CHROMATA online platform [3], which offers AI tools for analyzing, classifying, and annotating ICH content. This integration assists designers in creating immersive experiences based on ICH.