基于蒙皮模型和WEOPA的人体外形和运动跟踪

为了解决多关节人体三维外形和运动跟踪问题,提出了一种基于蒙皮模型(skinned model)和加权扩展正交普鲁克分析(weighted extended orthogonal procrustes analysis,WEOPA)的三维人体外形和运动跟踪方法。首先使用Shape-From-Silhouettes方法从当前帧多目视频图像轮廓中计算出被跟踪人物的可视外壳,然后建立人物先验蒙皮模型并与之匹配,其中蒙皮模型的引入使得多关节人体三维外形和运动跟踪问题被转化为一个分层非线性最小二乘匹配问题。考虑到在先验模型获取中肢体间连接部位产生的网格顶点粘连问题,采用WEOPA算法来进行求解,其中通过引入4种不同的权值分配策略,克服了顶点粘连对匹配的影响,并有效利用了匹配点对的几何法线信息。实验结果表明,该方法能对人物外形和运动进行有效的跟踪。

Human body shape and motion tracking using skinned model and WEOPA

A new 3D human body shape and motion tracking method based on skinned model and WEOPA(weighted extended orthogonal procrustes analysis) is proposed.First,the visual hull of the tracked people in current frame is computed using Shape-From-Silhouettes method from multi-view silhouettes of the synchronized videos;and then,a prior skinned model is created and fitted to the visual hull.Based on the skinned model,the 3D human body shape and motion tracking problem can be transformed into a hierarchical nonlinear least squares matching problem.In order to solve the mesh vertex adhesion problem between limbs and torso of the prior human mesh model,WEOPA algorithm is adopted.Using 4 different weight assigning methods,WEOPA can overcome the bad influence of the mesh vertex adhesion and use the geometric normal information of vertices during fitting.Experimental results show that the proposed method can effectively conduct 3D human body shape and motion tracking.