Enhancing Dance Performance for Body Motion Interaction Through Swarm Intelligence and Deep Learning

This paper proposes a novel framework to combine deep learning-based pose estimation with Swarm Intelligence (SI) for real-time stage optimization. In details, a multi-stage convolutional neural network is first employed to extract accurate skeletal keypoints from live video feeds of dance performances, while temporal smoothing techniques mitigate noise and occlusion in complex choreographic scenarios. These pose data are then used by an SI optimizer to iteratively refine stage parameters—lighting color, brightness, projection content—under stringent low-latency requirements. By formulating an objective function that captures aesthetic, rhythmic, and latency considerations, the swarm adaptively converges on stage configurations that maintain strong synchronization with the dancer's motions and preserve artistic coherence. Extensive experiments on both curated datasets and real-stage recordings demonstrate that our method achieves substantially higher synchronization and aesthetic quality scores than static or rule-based approaches, while incurring manageable computational overhead.