Computer vision technology and video image processing technology in the visual rehearsal of sports dance is a hot research topic. Based on this research background, the thesis uses 3D mathematical modelling technology to interpolate and extract the captured sports and dance movement information to make the final synthesised human animation natural, smooth and lifelike. At the same time, the thesis realises the method of action cohesion through the definition of characteristic action unit attributes and association constraints. Then, it applies it to the visual rehearsal system of sports dance. Finally, the analysis of experimental results proves that the proposed method can improve the precision and recall of rehearsal.