基于快速聚合和预配准的三维人脸重建  

3D Face Reconstruction Based on Fast Aggregation and Pre-registration

作　　者：刘建斌 马燕[1] 黄慧 LIU Jianbin;MA Yan;HUANG Hui(College of Information,Mechanical and Electrical Engineering,Shanghai Normal University,Shanghai 200234)

机构地区：[1]上海师范大学信息与机电工程学院,上海200234

出　　处：《计算机与数字工程》2024年第5期1477-1481,1509,共6页Computer & Digital Engineering

基　　金：国家自然科学基金项目(编号:61373004)资助。

摘　　要：为提高三维人脸重建的时间效率,提出一种基于快速聚合和预配准的三维重建方法。首先,使用两个普通摄像头构建双目立体视觉系统,通过标定双目摄像头获取相机参数和人脸图像对,并进行图像对校正、人脸特征点检测和前景提取。其次,根据人脸特征点计算合适的视差搜索范围,从而避免手动设置;在不影响精度的前提下,使用快速代价聚合方法提升聚合的时间效率;采用多角度孔洞填充方法对孔洞进行填充,从而得到较平滑的视差图。接着,利用基于人脸特征的点云配准从恢复的三维点云中提取标记的人脸特征点并进行配准,从而提高配准的精度和效率。最后,使用Possion重建方法进行人脸表面恢复。实验结果表明,改进后方法的平均时间效率提高了26.45%,时间效率高并且重建出的三维人脸轮廓真实、结构清晰。In order to improve the time efficiency of 3D face reconstruction,a 3D reconstruction method based on fast aggrega-tion and pre-registration is proposed.Firstly,binocular stereo vision system is constructed with two ordinary cameras.The camera parameters and face image pair are acquired by calibrating the cameras,and image pair correction,facial feature point detection and foreground extraction are conducted.Secondly,the appropriate disparity search range is obtained according to the facial feature points,which can avoid the manual setting.The fast cost aggregation method is used to improve the efficiency in the assurance of ac-curacy.The holds are filled with a multi-angle hold filling algorithm,which can obtain smoother disparity map.Next,the facial fea-ture points are extracted and registered from the restored 3D point cloud based on the facial feature point cloud registration,which can further improve the accuracy and efficiency of the registration.Finally,the Possion reconstruction method is used to recover the face surface.The experimental results show that the average time efficiency of the improved method is increased by 26.45%,the time efficiency is high,and the reconstructed 3D face contour is real and the structure is clear.

关 键 词：双目立体视觉 三维重建 人脸特征点 点云配准 Possion重建 

分 类 号：R339.14[医药卫生—人体生理学]