A Dance Emotion Recognition Method Based on Movement Analysis and Federated Learning in Edge-Cloud Hybrid Environments

With the development of dance diversity, substantial volumes of big data associated with dance movements have been produced, predominantly in the form of videos and images. Dance uses movements to convey emotions, which is considered the essence of choreographic works. Movements used to express an inner emotional manifestation are the external factors that constitute dance, visible, and tangible actions. However, for online education, identifying dance emotions from the dance movements to support the grading of students' dance has become an urgent challenge. In this paper, we first construct a cloud-edge hybrid environment to be used for dance learning. Then, we propose a novel approach for expressing dance emotions by utilizing advanced artificial intelligence techniques to encode different dance movements and emotions. Leveraging Laban Movement Analysis theory, we employ a federated learning framework in the hybrid environment to train dancer movements and guarantee the privacy of dancers. This process yields feature parameters related to the dancer's body structure, spatial orientation, and force dynamics. The original dance motion data is transformed into these three feature representation parameters, resulting in dance emotion data. The experimental results demonstrate that recognizing dance emotions from the perspective of dance movements is feasible, the accuracy increases by 4.22% than other state-of-the-art methods.