三维动态脚型测量方法

Acquisition of time-varying 3D foot shapes from video

基于多视几何原理来恢复和重建随时间变化的三维柔性物体是当前计算机视觉和三维扫描领域的研究热点之一.本文利用主动式标记点方法,提出并实现了在三维模型指导下鲁棒地从多视点视频中恢复和重建三维脚型的方法.它首先基于传统立体视觉方法重建出参考帧中的三维脚模型,然后在此基础上,通过构建三维模型顶点运动投影的速度矢量场来建立相邻帧之间的标记点对应关系,鲁棒地跟踪复杂背景中的三维动态目标以及相邻帧间运动幅度较大的三维动态目标.而对于在运动过程中由于遮挡等原因所造成的标记点缺失,则在三维参考模型的指导下,利用数字几何处理方法可靠地在三维空间中估计出相应的标记点,从而有效保证了三维脚模型在运动过程中的完整性.实验结果表明:该方法可以鲁棒、可靠地获得动态的三维脚型.

The 3D scanning of time-varying objects based on multi-view geometry has been a hot topic in computer vision in recent years. This paper presents an active approach to acquiring 3D shapes of moving foot from multi-view video sequences based on the explicit marker points woven in socks. The reference 3D model of foot is firstly recovered from the starting frames in the multi-view video clips, and then the markers in each view are traced as a whole based on a continuous motion vector field built from the reference 3D model. The missing vertices in the candidate 3D model caused by self-occlusion, non-uniform illumination and random noises are reliably estimated and reconstructed in 3D space by minimizing the changes of differential features in 3D geometry of the reference model. The experimental results show that our method can robustly recover the 3D model of foot in complex background even if there is a relatively large movement between the adjacent frames, with an acceptable accuracy of the resulting 3D model.