自由视角三维扫描数据拼合技术研究

发布时间：2019-03-20 14:47

【摘要】：非接触式三维测量是人类认识世界手段的一次巨大飞跃,其中基于光学的三维扫描设备大大提高了人们测量的便捷性。由于被测物自身的遮挡以及三维扫描设备视场的限制,一般需要将经由扫描获得的多视角点云数据进行拼接融合后方能获得对被测物三维信息的完整描述。目前三维扫描设备单视角精度已经能达到很高的精度,因此,点云的拼接融合对最终测量结果的影响很大。本文基于自由视角结构光三维扫描,对点云的拼接融合相关技术进行了研究,主要研究内容及成果如下:(1)为解决结构光三维扫描单视角测量的局限性,设计了一种自由视角三维测量系统,提出了一种改进的高稳定映射拼接方法。根据图像间匹配的特征点实现视角重合,利用结构光相位信息完成特征点三维映射和空间定位。针对特征点相位缺失无法重建的普遍问题,利用邻域点插值恢复特征点方法,虚拟重建视角间有效映射点,通过四元数法实现了视角间点云的刚性变换和拼接。(2)提出了一种基于图像k邻域的快速法矢估算方法,在保证估算的法矢精度前提下,大大提高了法矢估算的速度。基于估算获得的法矢,利用ICP算法对点云进行精拼接,进一步提升了经由映射拼接进行初步定位后的两视角点云间的拼接精度。(3)研究了一种多视角点云拼接方法。为描述多视角点云间的关系,引入图论中最小生成树的概念,基于最小生成树,构建环形结构。针对环形结构点云拼接遇到的问题,设计了一种新的拼接模型对环形结构拼接中产生的累积误差进行分配。对于拼接完成的多视角点云,进行了基于GPU的点云融合,大大提高了点云融合的速度。(4)搭建了实验平台,通过实验对相关算法的有效性和实用性进行了验证。
[Abstract]:Non-contact three-dimensional measurement is a great leap in human understanding of the world, in which the optical-based three-dimensional scanning equipment greatly improves the convenience of people's measurement. Due to the occlusion of the measured object and the limitation of the field of view of the three-dimensional scanning equipment, it is generally necessary to splice and fuse the multi-view point cloud data obtained by the scanning to obtain a complete description of the three-dimensional information of the measured object. At present, the precision of single view angle of 3D scanning equipment can reach very high precision, so the splicing fusion of point cloud has a great influence on the final measurement results. Based on the 3D scanning of structured light with free view, this paper studies the fusion technology of point cloud splicing. The main contents and achievements are as follows: (1) in order to solve the limitation of single angle measurement of structured light 3D scanning, the main research contents and achievements are as follows: (1) A free angle 3D measurement system is designed and an improved high stability mapping splicing method is proposed. According to the matching feature points between images, the angle of view is coincident, and the structural light phase information is used to complete the three-dimensional mapping and spatial localization of the feature points. In order to solve the general problem that the phase loss of feature points cannot be reconstructed, the feature points can be recovered by neighborhood interpolation, and the effective mapping points between different angles of view can be reconstructed by virtual reconstruction. Through the quaternion method, the rigid transformation and splicing of the point cloud between the angles of view are realized. (2) A fast normal vector estimation method based on image k-neighborhood is proposed, which greatly improves the speed of normal vector estimation on the premise of guaranteeing the accuracy of normal vector estimation. Based on the estimated normal vector, the ICP algorithm is used to concatenate the point clouds, which further improves the stitching accuracy of the two-view corner clouds after the initial positioning by mapping splicing. (3) A multi-view point cloud splicing method is studied. In order to describe the relationship between multi-view point clouds, the concept of minimum spanning tree in graph theory is introduced, and the ring structure is constructed based on the minimum spanning tree. Aiming at the problems encountered in point cloud splicing of ring structure, a new splicing model is designed to allocate the accumulated errors in ring structure splicing. For the multi-view point cloud, the point cloud fusion based on GPU is carried out, which greatly improves the speed of point cloud fusion. (4) the experimental platform is built, and the validity and practicability of the related algorithms are verified by experiments.
【学位授予单位】：南京航空航天大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP391.41

【相似文献】