基于摄像机阵列的大范围三维场景重建

发布时间：2018-01-06 10:37

本文关键词：基于摄像机阵列的大范围三维场景重建　出处：《西南科技大学》2017年硕士论文　论文类型：学位论文

【摘要】：基于视觉处理的三维信息获取技术已经成为计算机视觉领域非常重要的研究方向。目前常用的获取场景三维信息的方法有基于双目立体视觉、结构光编码以及TOF深度相机等。然而这些三维获取技术不能很好的满足场景的多方位、高精度、快速的三维重建要求。针对此问题,本课题在室内大视场环境下搭建摄像机阵列三维重建系统,在双目立体视觉的基础上使用摄像机阵列三维场景重建方法,通过拼接各个视角下双目重建获得的点云,得到大范围多视角全景三维模型。本课题主要研究内容包括摄像机阵列的标定,快速双目立体匹配和三维重建,摄像机阵列的全局拼接以及摄像机阵列的软硬件系统实现。首先,在双目摄像机标定的基础上,本文设计了一种灵活的摄像机阵列标定方法,实现对阵列摄像机的内部参数和外部位姿参数的标定。实验证明该标定方法可以实现摄像机阵列的高精度全局标定,双目标定精度达到0.0822mm,摄像机阵列标定参数达到0.079mm。其次,针对现有的双目匹配算法计算量大,速度慢等缺点,本文将SIFT算法的实现转移到图形处理器(GPU)上加快稀疏特征点的提取速度,并且在此基础上使用基于贝叶斯模型的稠密匹配算法实现双目快速稠密匹配。然后,本文使用摄像机阵列外部标定参数初步对齐各个视角下的点云,再利用ICP(Iterative Closest Point)算法优化点云拼接精度,完成高精度三维场景完整重建的目的。最后,本文搭建了摄像机阵列硬件平台和软件系统。硬件平台实现了摄像机阵列的信号控制、同步相机触发以及高带宽图像数据传输。软件系统将摄像机阵列重建算法集成在一起,实现有效的人机交互和高效的可视化效果。
[Abstract]:3D information acquisition technology based on visual processing has become a very important research direction in the field of computer vision. At present, the commonly used methods to obtain 3D information of scene are based on binocular stereo vision. Structured light coding and TOF depth camera, however, these 3D acquisition techniques can not meet the multi-directional, high-precision, fast 3D reconstruction requirements. In this paper, the camera array 3D reconstruction system is built in the indoor large field of view environment, and the camera array 3D scene reconstruction method is used on the basis of binocular stereo vision. By splicing the point clouds obtained by binocular reconstruction in each angle, the panoramic 3D model of a wide range and multi-view angle is obtained. The main contents of this research include camera array calibration, fast binocular stereo matching and 3D reconstruction. The global splicing of the camera array and the realization of the hardware and software system of the camera array. Firstly, based on the binocular camera calibration, a flexible camera array calibration method is designed in this paper. The calibration of the internal and external pose parameters of the array camera is realized. The experiment shows that the calibration method can achieve the high precision global calibration of the camera array, and the precision of double target determination can reach 0.0822 mm. The calibration parameters of camera array are 0.079mm. secondly, the existing binocular matching algorithm has the disadvantages of large computation and slow speed. In this paper, the implementation of SIFT algorithm is transferred to graphics processor (GPU) to speed up the extraction of sparse feature points. On this basis, the dense matching algorithm based on Bayesian model is used to realize the binocular fast dense matching. Then, the external calibration parameters of camera array are used to preliminarily align the point clouds in each angle of view. Then using ICP(Iterative Closest Point algorithm to optimize the point cloud stitching accuracy, complete the goal of complete reconstruction of high-precision 3D scene. Finally. In this paper, the hardware platform and software system of the camera array are built, and the signal control of the camera array is realized by the hardware platform. The software system integrates camera array reconstruction algorithm together to achieve effective man-machine interaction and efficient visualization.
【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】