双目立体视觉的三维人脸重建方法

3D face reconstruction using binocular stereo vision

创建逼真的三维人脸模型始终是一个极具挑战性的课题.随着三维人脸模型在虚拟现实、视频监控、三维动画、人脸识别等领域的广泛应用,三维人脸重建成为计算机图像学和计算机视觉领域的一个研究热点.针对这一问题,提出一种基于双目立体视觉的三维人脸重建方法,重建过程中无需三维激光扫描仪和通用人脸模型.首先利用标定的2台摄像机获取人脸正面图像对,通过图像校正使图像对的极线对齐并且补偿摄像机镜头的畸变;在立体匹配方面,选择具有准确可靠视差的人脸边缘特征点作为种子像素,以种子像素的视差作为区域生长的视差,在外极线约束、单调性约束以及对应匹配的边缘特征点的约束下,进行水平扫描线上的区域生长,从而得到整个人脸区域的视差图,提高了对应点匹配的速度和准确度;最后,根据摄像机标定结果和立体匹配生成的视差图计算人脸空间散乱点的三维坐标,对人脸的三维点云进行三角剖分、网格细分和光顺处理.实验结果表明,该方法能够生成光滑、逼真的三维人脸模型,证明了该算法的有效性.

Generation of lifelike 3D human faces is a challenging task.Recent increases in the use of 3D face models in virtual reality,video surveillance,3D animation,and face recognition have led to 3D face reconstruction becoming a research hotspot.The authors proposed a 3D face reconstruction method based on binocular stereo vision theory.After capturing the front view of a face with two calibrated cameras,the captured pair of stereo images were rectified to align their epipolar lines and compensate for image distortions.To obtain maps with accurate matching and dense disparity,a stereo matching algorithm based on region growing was developed.An edge feature point with reliable disparity was selected as a seed point.Then region growing was performed along horizontal scan-lines under multi-constraints.With camera calibration and disparity mapping,3D coordinates of corresponding points were calculated.Then the face model was reconstructed with a series of methods,such as Delaunay triangulation,mesh subdivision and smoothing.Experimental results showed that the method can generate a smooth and lifelike 3D face model.