Creating a dataset of rumba dance videos for an application that provides feedback to dancers

This bachelor thesis focuses on preparing a dataset of rumba dance videos for the application of neural networks in the analysis and provision of feedback to dancers. The aim of the thesis is to generate a comprehensive and diverse dataset of dance videos, which will serve for training and testing neural network models. The dataset is created from publicly available sources, such as YouTube, and further supplemented with videos from recordings conducted under specified conditions. Data processing includes the extraction of key points of dancers and the analysis of movement using technologies such as OpenPose. The qualitative evaluation of the dataset focuses on the accuracy of key element extraction and the overall quality of the videos. The thesis also includes a comparison of the created dataset with existing data in the field of dance analysis. Key aspects of the thesis are illustrated based on a library of references containing articles and books relevant to the issue. The result of the thesis is a high-quality dataset that can be used for the development of an application providing feedback to dancers from video.