A multi-person human model reconstruction method based on latent space motion coding

Abstract

The invention discloses a multi-person human body model reconstruction method based on latent space motion coding, which specifically includes steps: S1. Hidden space motion prior training: S2. Building a multi-view camera system and collecting multi-person motion videos; S3. Steps The multi-body motion video collected by the multi-view camera system built in S2 is preprocessed using the existing open source 2D pose estimation and tracking to obtain the 2D joint point trajectory; S4. Obtain the initial camera external parameters; S5. Complete the multi-body motion Alignment of each human body to be reconstructed in the video; S6. Iteratively optimize the latent space encoding and the initial camera external parameters obtained in step S4, so that the joint point projection of the three-dimensional human body model is consistent with the two-dimensional joint point trajectory obtained in step S3, and finally Skinning the mannequin in each frame yields a human reconstruction for the entire sequence. The invention reduces the influence of multiple people's mutual occlusion and tracking error on reconstruction.

Description

technical field [0001] The invention relates to a multi-person human body model reconstruction method based on latent space motion coding, which belongs to the field of computer vision and three-dimensional vision. Background technique [0002] Multi-person human body model reconstruction plays an important role in applications such as holographic communication, traffic behavior monitoring, and group behavior analysis. Driven by artificial intelligence-related technologies, the market demand for human body reconstruction is increasing. At present, the human body model reconstruction scheme based on single-view is faced with problems such as depth ambiguity and unrobust to occlusion. Missing, in the face of occluded multi-person images, good reconstruction results cannot be obtained. Although the human body model reconstruction scheme based on multi-view can make up for these shortcomings of the single-view scheme to a certain extent, the existing methods need to perform ca...

AI Technical Summary

Problems solved by technology

However, the reconstruction method with timing constraints has a large solution space dimension, complex calculations, and is easy to fall into a local minimum

Therefore, on the basis of simplifying the reconstruction process, it is a new challenging problem to add timing information to achieve accurate multi-person human body model reconstruction.

Images