Mas multi-view action megre 3D

Abstract

In this experiment, the 2D human skeleton is detected from various viewpoints by reading in the data of athletes in 2D motion; then the geometric relationship is used to calculate the distance and match the human body from multiple viewpoints; finally, the 3D coordinates are calculated by using the 2D coordinates of the multiple viewpoints and the parameters of the camera, and the diffusion model trained on the 2D data using the MAS only requires multiple sets of different viewpoints to faster and more accurately build up the human body's motion image without any dead angle. The human motion image is constructed without any dead angle by using the diffusion model trained by MAS on 2D data