Deep Learning for Dance Teaching System

Abstract

In the aspect of virtual human animation technology, the traditional way is that teachers manually mark the actions frame by frame, which results in heavy workload, high cost of motion capture and insufficient accuracy of manual recognition. In this paper, linear interpolation and quaternion spherical interpolation are combined to interpolate the captured motion information, which solves the problem of jumping between animation frames caused by a single interpolation algorithm, and makes the final human animation natural, smooth and realistic. In this paper, the precise data analysis of each rigid body motion segment is done. The key frame spline interpolation algorithm is used to solve the position offset problem, and the quaternion interpolation algorithm is used to solve the rotation problem of the action body or the action limb, so that the action modification made by the computer-aided action arrangement system can completely return to the virtual human animation.