摘要
为了从图像序列中恢复牙齿的三维结构,对从运动中恢复(SFM)和多视图面片(PMVS)三维重建方法进行了研究;首先,利用SFM方法从图像序列中恢复相机参数并估计相机位置;其次,针对标定好的序列图像,通过Harris和DOG检测特征点并在图像对中匹配,得到一系列稀疏的面片,根据光照一致性扩展这些初始的匹配到邻近像素,得到比较密集的面片,然后利用可视化约束条件,消除错误的匹配,最终生成三维模型的面片集合;最后把生成的面片集合转换为点云集合;经过大量的实验得出了一组具有较好实验效果的参数,分别为β1=2,β2=16,μ=5,α0=0.5;实验结果表明,该算法能够有效地重建出牙齿的三维结构,并具有很好的视觉效果;SFM方法能够有效地标定相机,基于多视图面片三维重建方法能够很好地重建出物体的三维模型,两种方法相结合是非常好的三维重建方法。
In order to reconstruct the 3D structure of the teeth model from the image sequence, the structure from motion （SFM） and Patches-Based Multi-View Stereo （PMVS） method was studied. Firstly, the SFM method was used to recover the camera parameters and estimation of the position of the camera calibration; secondly, according to the calibrated image sequences, use the Harris and DOG detect feature points in each image and matching the points in image pair, get a series of sparse patches and use the photometric to extent these ini- tial matching to neighboring pixels to get relatively dense patches, then use the visualization constraint condition to eliminate the error matc- hing, generate a 3D model of the patch set; finally, convert the patch set to point cloud collection. The experimental results show that the al- gorithm can effectively reconstruct the 3D structure of the teeth model, and it has a very good visual effect. The SFM method can calibrate the camera effectively, The Patches-Based Multi-View Stereo method can reconstruct the 3D model effectively, so the combination of these two methods is a very good method for 3D reconstruction.
出处
《计算机测量与控制》
北大核心
2013年第4期1067-1070,共4页
Computer Measurement &Control
基金
中国科学院知识创新工程重要方向项目(KGCX2-YW-909)
苏州生物医学工程技术研究所二期建设重大项目(PET/CT项目)(Y053011305)
苏州市科技计划项目资助(YJS0952)
长春市科技计划项目资助(09K218)
