基于SMPL模型的人体姿态估计  

Human Pose Estimation Based on SMPL Model

作　　者：李健[1] 杨镖镖 张皓若 LI Jian;YANG Biao-biao;ZHANG Hao-ruo(College of Electronic Information and Artificial Intelligence,Shaanxi University of Science and Technology,Xi'an Shanxi 710021,China)

机构地区：[1]陕西科技大学电子信息与人工智能学院,陕西西安710021

出　　处：《计算机仿真》2021年第3期292-297,486,共7页Computer Simulation

基　　金：国家土建结构预制装配化工程技术研究中心开放课题资助(2019CPCCE-K02);国家重点研发计划资助(2018YFB1305300);国家自然科学基金项目(61825303,U1713215);陕西省工业攻关项目(2015GY044)。

摘　　要：针对目前人体形变模型中姿态估计算法容易出现误差、信息缺失等问题,提出一种利用深度相机获取的人体三维信息来优化模型的方法。通过深度相机Kinect获取的三维骨架信息,与SMPL模型进行配准,修正原始的模型姿态,得到一个接近人体真实姿态的模型。实验结果表明,融合人体三维信息后,模型的准确性得到一定程度上的提高。Aiming at the current pose estimation algorithms in human body deformation models that are prone to errors and lack of information,a method for optimizing the model is proposed,using the three-dimensional information of the human body obtained by the depth camera.The three-dimensional skeleton information obtained by the depth camera Kinect was registered with the SMPL model,and the original model pose was corrected to obtain a model close to the real pose of the human body.The experimental results show that the accuracy of the model is improved to a certain extent after fusing the three-dimensional information of the human body.

关 键 词：形变模型 姿态估计 深度相机 三维骨架 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]