基于点云处理的场景三维重建

发布时间：2019-02-21 15:27

【摘要】：随着获取点云设备的发展,人们能够方便快捷地获得实际场景的大量点云数据,促进了逆向工程技术的发展。在三维重建的过程中,为了获得大范围场景的点云,需要从不同的角度去获取场景的点云数据,所以点云的拼接和重建技术显得尤为重要。论文对点云数据处理的拼接和重建算法以及相关问题进行研究。针对点云拼接中存在耗时长、拼接效果差等问题,本文研究了现有的ICP拼接算法以及LM、SVD获得变换矩阵的方法。通过利用正交映射,获取三维点云与二维图像之间的映射关系,并在二维图像中获取匹配点对,提出采用LM和SVD结合的方法来获取变换矩阵,实现点云的拼接。最后将几种拼接算法的实验结果进行分析比较,验证了本文提出的改进算法,在极大缩短拼接时间的同时,降低了拼接误差。针对点云重建中存在的抗噪性差、耗时长、不光滑且存在孔洞等问题,本文研究了泊松算法和贪婪投影三角化算法。然后在贪婪投影三角化算法的基础上,首先,提出一种改进的加权移动最小二乘方法来进行法线估计,使得重建效果在平滑的同时,具有了抗噪性;其次,提出了改进点云预处理方法,在平滑的基础上进行了数据重采样,使得重建效果更加光滑且细化了孔洞;最后,针对重建耗时长的问题,使用八叉树替代KD树,在不影响重建效果的基础上,实现了重建时效的优化。
[Abstract]:With the development of point cloud acquisition devices, people can easily and quickly obtain a large number of point cloud data of the actual scene, which promotes the development of reverse engineering technology. In the process of 3D reconstruction, in order to obtain the point cloud of a large scale scene, it is necessary to obtain the point cloud data from different angles, so the point cloud splicing and reconstruction technology is particularly important. In this paper, the algorithms and related problems of point cloud data processing are studied. In order to solve the problems of point cloud stitching, such as time consuming and poor effect, the existing ICP stitching algorithm and the method of obtaining transformation matrix by LM,SVD are studied in this paper. By using orthogonal mapping, the mapping relationship between 3D point cloud and two-dimensional image is obtained, and matching point pairs are obtained in two-dimensional image. The method of combining LM and SVD to obtain transformation matrix is proposed to realize the splicing of point cloud. Finally, the experimental results of several stitching algorithms are analyzed and compared to verify the proposed improved algorithm, which can greatly shorten the stitching time and reduce the stitching error. In this paper, Poisson algorithm and greedy projection triangulation algorithm are studied to solve the problems of poor noise resistance, long time consumption, not smooth and holes in point cloud reconstruction. Then, based on the greedy projection triangulation algorithm, an improved weighted moving least square method is proposed to estimate the normal line, which makes the reconstruction effect smooth and noise resistant. Secondly, an improved point cloud preprocessing method is proposed, and the data resampling is carried out on the basis of smoothing, which makes the reconstruction effect smoother and finer. Finally, octree is used instead of KD tree to solve the problem of time-consuming reconstruction, and the reconstruction time is optimized without affecting the reconstruction effect.
【学位授予单位】：西南科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】