基于双目视觉的三维重建方法研究

发布时间：2018-06-04 00:47

本文选题：双目视觉 + 摄像机标定　；参考：《沈阳工业大学》2016年硕士论文

【摘要】：随着计算机科学领域的快速发展,机器视觉代替人眼去完成对物体的判断和测量并实现物体信息的三维重建已经成为了一个热门的研究方向。基于双目视觉的三维重建方法研究是利用两个摄像机从不同的角度对物体进行观察,利用立体匹配的结果计算出图像特征点位置的偏差,最终根据三角测量原理获取物体的深度信息,实现三维重建。其基本流程主要包括图像采集、摄像机标定、图像预处理、立体匹配、三维重建。本文首先完成了双目视觉的硬件平台搭建,介绍了双目视觉的数学模型。在图像采集方面,利用已搭建好的视觉平台并充分考虑了光照强度和目标物体特征等情况下完成了多组图像采集。在摄像机标定方面,根据非线性摄像机的成像模型原理,利用张正友平面模板标定方法计算出了摄像机的内参数、外参数以及畸变系数,为后续的处理做好了准备。在图像平滑方面,重点介绍了高斯滤波法和中值滤波法,并对图像平滑和中值滤波进行了实验对比;在边缘检测方面,对Roberts算子、Prewitt算子、Sobel算子、Log算子、Canny算子做了分析,并通过了实验进行了对比。在特征点提取方面,本文对传统的SURF算法进行了改进,主要通过对传统SURF算法检测到的特征点进行信息量约束和距离约束,去掉了一些信息含量较少、距离相对较近的特征点,使得提取的特征点个数适中,消除了特征点聚集现象。在特征点匹配方面,根据改进的SURF算法提取的特征点采用近邻匹配算法完成匹配,通过对三类不同的目标物体在两种不同条件下与传统SURF算法的对比结果表明,本文改进的SURF算法对亮度的变化具有一定的鲁棒性,匹配准确度和匹配所需要的时间优于传统的SURF算法,在检测到大量特征点的同时,能够有效的去掉一些信息量较少以及容易产生聚集现象的特征点。在三维重建方面,根据三维重建原理,计算得到物体的三维坐标,并最终生成了物体在三维空间中的深度图。
[Abstract]:With the rapid development of computer science, machine vision has become a hot research direction to complete the judgment and measurement of objects and realize the 3D reconstruction of object information instead of human eyes. The method of 3D reconstruction based on binocular vision is to use two cameras to observe objects from different angles, and to calculate the deviation of image feature points by using the results of stereo matching. Finally, the depth information of the object is obtained according to the triangulation principle, and the 3D reconstruction is realized. Its basic flow mainly includes image acquisition, camera calibration, image preprocessing, stereo matching, and 3D reconstruction. Firstly, the hardware platform of binocular vision is built, and the mathematical model of binocular vision is introduced. In the aspect of image acquisition, several groups of images are collected by using the established visual platform and fully considering the intensity of illumination and the characteristics of the target object. In the aspect of camera calibration, according to the principle of nonlinear camera imaging model, the camera's internal parameters, external parameters and distortion coefficient are calculated by using the calibration method of Zhang Zhengyou plane template, so as to prepare for the subsequent processing. In the aspect of image smoothing, the Gao Si filter and median filter are introduced in detail, and the image smoothing and median filtering are compared experimentally. In the aspect of edge detection, the Roberts operator Prewitt operator and the Roberts operator are analyzed. The experimental results are compared. In the aspect of feature point extraction, this paper improves the traditional SURF algorithm, mainly through the information constraint and distance constraint of the feature points detected by the traditional SURF algorithm, and removes some feature points which have less information content and are relatively close to each other. The number of feature points extracted is moderate and the phenomenon of feature point aggregation is eliminated. In the aspect of feature point matching, the nearest neighbor matching algorithm is used to match the feature points extracted from the improved SURF algorithm. The comparison of three kinds of target objects under two different conditions with the traditional SURF algorithm shows that, The improved SURF algorithm is robust to the change of luminance, and the matching accuracy and matching time are better than the traditional SURF algorithm. At the same time, a large number of feature points are detected. Can effectively remove some of the less information and easy to produce clustering feature points. In the aspect of 3D reconstruction, according to the principle of 3D reconstruction, the 3D coordinates of the object are calculated, and finally the depth map of the object in three dimensional space is generated.
【学位授予单位】：沈阳工业大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【相似文献】