Application of Deep Learning in Multi-Style Dance and Music Matching Choreography

Abstract

In recent years, with the development of the digital era, music to dance generation research has received extensive attention from industry and academia, and has become one of the basic tasks in the cross-modal field, which can be widely used in entertainment, education, virtual and other fields, and has a good application prospect. Based on deep learning, the choreography system is proposed in this paper. In this paper, a novel automatic music choreography system is proposed, aiming to realize efficient matching and generation of music and dance movements through deep learning models. In the study, a bidirectional cyclic gating unit is used as the core network structure, and multi-temporal modeling is carried out for the timing characteristics of the dance movements. Two standard motion capture datasets, Laban-16 and Laban-48, were used in the experiments to verify the effectiveness of the proposed algorithm. The accuracy of the model in continuous motion recognition is significantly improved on the Laban-16 dataset, with a recognition rate of 72.79%, demonstrating strong generative ability and innovation, and the recognition rate on the Laban-48 dataset is 68.92%. In addition, the experimental results show that the multi-temporal-based algorithm outperforms the traditional joint features in motion recognition performance, verifying the effectiveness of feature selection.