Image-based smpl parameter prediction and human body model generation method

Abstract

The invention discloses a picture-based SMPL parameter prediction and human body model generation method, and relates to the fields of machine learning, computer vision and three-dimensional reconstruction. The present invention includes: predicting the morphological parameters of SMPL, extracting the human body outline from the input image through a deep neural network with feature fusion and attention mechanism, and then predicting the morphological parameters of the picture of the human body outline, and based on the established human body shape and morphological parameters. The mapping function is fine-tuned; the multi-stage attitude parameter prediction network is used to predict the attitude parameters of SMPL for the input image, and the 2d joint point coordinates, camera parameters and 3d joint point coordinates are used to train and predict the network; the final 3D is obtained through the SMPL conversion function mannequin. The present invention makes full use of camera information and 3D information, improves the effect of human body contour extraction and posture parameter prediction, improves the fitting effect of a three-dimensional model to human body shape in pictures, and has universal applicability.

Description

technical field [0001] The invention relates to the fields of machine learning, computer vision and three-dimensional reconstruction, in particular to a picture-based SMPL (Skinned Multi-Person Linear) parameter prediction and human body model generation method. Background technique [0002] 3D models are now widely used in many fields such as virtual reality, 3D games, and virtual fittings. At present, large-scale 3D human body modeling still relies on hardware devices such as lidar and depth cameras. The main disadvantage of this method is that it is limited to indoor environment, cannot get rid of the shackles of clumsy hardware devices in reality, and is not portable. There are also some methods that choose to use the depth camera as the image acquisition device, use the depth camera to take pictures of the modeling object, and combine the SMPL model with the point cloud image to achieve the purpose of 3D human body modeling. Compared with the scanning scheme, this kind...

AI Technical Summary

Problems solved by technology

[0006] The present invention aims at the use of deep learning technology to build a three-dimensional model of the human body, the application of 3D information in the collected pictures is less, the research on SMPL parameters is lacking, the generation effect of the three-dimensional model of the human body cannot be adjusted, and the complex and changeable real scene cannot be adapted, etc. problem, providing a picture-based SMPL parameter prediction and human body model generation method

Images