A Method for Pose and Shape Reconstruction of 3D Human Model from Video Sequence

Abstract

The invention discloses a three-dimensional human body model attitude and shape reconstruction method through a video sequence, and the method comprises the steps: 1), extracting two-dimensional jointpoints of a person in a video frame; 2), extracting facial feature points of the person in the video frame; 3), extracting a two-dimensional contour line of the person in the video frame; 4), estimating camera parameters, and generating a perspective matrix according to the camera parameters; 5), carrying out the optimization of the model posture and shape through an optimization method; 6), obtaining the edge points of the 3D human body model at the camera angle and the matching relationship between the edge point and the 2D contour line; 7), carrying out the re-optimization of the shape ofthe model through the optimization method so that the sum of the differences between the distances from the positions of the edge points of the three-dimensional model at the camera angle after projection to the positions of the two-dimensional contour points in the video frame is minimized. The method can avoid the posture turnover in the reconstruction process, and also can accurately reconstruct the posture of a head under the condition that the overall similarity of the model is guaranteed.

Description

technical field [0001] The invention relates to the technical field of computer graphics and three-dimensional human body mesh model reconstruction, in particular to a three-dimensional human body model posture and shape reconstruction method of video sequences. Background technique [0002] 3D human body model posture and shape reconstruction refers to obtaining information about human body movement through information collection, and then reconstructing a 3D human body model based on this information. This technology is of great significance for 3D human body modeling, model editing, game production, etc. The pose and shape reconstruction of the 3D mannequin of the video sequence is to extract the pose and shape information from the video of human motion, and use this information to reconstruct the motion sequence of the 3D mannequin with similar pose and shape to the video characters. [0003] At present, the pose and shape reconstruction methods of 3D human body models ...

AI Technical Summary

Problems solved by technology

However, limited by the configuration of special equipment, such methods cannot be well applied in real life.

The main disadvantages of the above-mentioned 3D human body model reconstruction method based on video or image are: (1) the pose flipping problem cannot be avoided; (2) the reconstructed model has a large difference in shape from the video image or only includes a skeleton model; (3) ) The reconstructed 3D model head state is inaccurate

Images