基于二维图像的参数化人体建模

Parametric Human Modeling Based on Two-Dimensional Image

为解决现有基于二维图像的三维人体建模容易出现误差、模型参数估计不准确等问题,结合深度学习和优化后的人体建模算法,并利用深度学习算法提供的结果作为先验,缓解优化算法对参数初始值敏感的问题,以实现针对单帧RGB图像和多帧RGB图像的三维人体建模。同时基于HMR深度学习模型预测SMPL-X模型的初始参数,通过添加人体轮廓、二维关键点等对模型参数进一步优化求解,并利用视频序列的帧间连贯性对视频帧中人物的三维姿态进行约束。结果表明:添加人体分割轮廓约束可使重建模型更加贴合人体,能够提升模型拟合的精度;基于视频序列帧间连贯性重建的人体模型,可减少身体旋转、四肢弯曲等误差出现的可能性,与目标对象姿态更加相近。

The existing 3D human modeling based on two-dimensional images is prone to errors and inaccurate estimation of model parameters.In order to alleviate the problem that the optimization algorithm is sensitive to the initial value of parameters,the results provided by the deep learning algorithm are used as a priori,and the human modeling algorithm based on deep learning and optimization is combined.This enables 3D human modeling for single frame RGB images and multi frame RGB images.The initial parameters of SMPL-X model are predicted based on HMR deep learning model,and the model parameters are further optimized by constraints such as human contour and two-dimensional key points.The inter frame coherence of the video sequence is used to constrain the 3D pose of the characters in the video frame.The results showed that adding human segmentation contour constraints could make the reconstructed model more fit the human body and improve the accuracy of model fitting.The pose of the human model reconstructed based on the inter frame coherence of the video sequence was more similar to that of the target object,which reduced the possibility of errors such as body rotation and limb bending.