一个基于多视图立体视觉的三维重建方法

摘要

A 3-D object reconstruction method based on multi-view stereo vision was studied.Given a set of images to reconstruct object,the depth map for each image was calculated by taking advantage of traditional stereo vision technology.Then a model fusion method based on voxel dividing was designed,merging the depth information of the image into a complete 3-D initial model,which was iteratively refined globally and output the final reconstruction model.When constructing voxel network diagram,a connection relationship between the voxels was established by making full use of images consistency measure.In iteration refinement to the model,the image consistency measure was combined with mesh smoothing criteria.The method handled the texture missing area very well and the complete 3-D model was successfully obtained.Experimental results showed that this method could reconstruct the high quality and complete 3-D model.%研究了基于多视图立体视觉的三维重建方法.给定一组待重建物体的图像,利用传统的立体视觉技术计算每幅图像的深度信息；然后设计一个基于体素划分的模型融合方法,将图像的深度信息融合为一个完整的三维初始模型；随后,对初始模型进行全局的迭代优化,输出最终的重建模型.在构造体素网络图时,充分利用图像一致性度量建立体素之间的连接关系；在对模型迭代优化时,结合图像一致性度量和网格的光顺性准则.该方法能很好地处理纹理缺失的区域,获得完整的三维模型.实验结果证明:该方法能够重建高质量完整的三维模型.