High-precision human body posture reconstruction method and system

Abstract

The invention discloses a high-precision human body posture reconstruction method and system, and the method comprises the following steps: obtaining a human body video, and selecting a key frame of the human body video; building a deep neural network to obtain human body contour information and two-dimensional joint information of the key frame; constructing a parameterized human body model, andconstraining the parameterized human body model by utilizing the human body contour information and the two-dimensional joint information to obtain initialized human body posture parameters; and optimizing and initializing human body posture parameters by utilizing the two-dimensional joint information and the human body contour information, selecting all frames of a human body video, constructinga joint constraint, optimizing and initializing the human body posture parameters again, and iterating two optimization processes by utilizing a Gauss-Newton formula until an energy difference valueof two adjacent iterations is smaller than a preset threshold value. The human body model can be better reconstructed by utilizing a few angles of human body RGB information, so that the reconstructedhuman body model more accurately reflects the real human body figure.

Description

Technical field [0001] The invention relates to the technical field of computer vision, in particular to a high-precision human body posture reconstruction method and system. Background technique [0002] With the continuous development of 3D reconstruction technology in the field of computer vision, the 3D reconstruction technology of the human body has become a research hotspot in the computer room field. In real life, there are a large number of pictures containing character information on the Internet. How to use these pictures to reconstruct the corresponding three-dimensional human body model has become an emerging research direction. [0003] The currently popular human body reconstruction methods based on RGB images are mainly divided into reconstruction of parametric human body models and reconstruction of human body models based on voxel representation. For the problem of RGB image reconstruction, the general solution is to first estimate the two-dimensional joint positi...

AI Technical Summary

Problems solved by technology

[0004] Although the technology of reconstructing 3D human body models based on RGB images has been greatly developed, the two popular methods still have their own problems: the reconstruction of parametric models results in limited ability to express the geometric details of the human body surface, and cannot be well represented. Reconstruct the detailed texture of human surface clothing

The human body model based on voxel representation is due to the limitation of the current video memory, the highest precision can only reach 128*128*128, the overall performance is relatively rough

Images