基于ARM的重启动随机游走立体匹配算法研究与实现

发布时间：2018-06-07 01:17

本文选题：双目视觉 + 立体匹配　；参考：《西南石油大学》2016年硕士论文

【摘要】：随着计算机视觉在近些年的飞速发展,双目立体视觉作为其重要研究方向之一,也得到了快速地发展,应用领域也不断拓宽。完整的双目视觉系统主要由图像采集、摄像机标定、图像校正、立体匹配和三维重建等步骤构成。而立体匹配作为其中最关键的步骤,匹配的精度和效率以及硬件实现直接决定了整个双目视觉系统的性能和实际应用。因此研究立体匹配算法及其实现对双目视觉的发展具有重大意义和实用价值。到目前为止,有许多不同种类的立体匹配算法相继被提出,但是立体匹配中存在的弱纹理,深度不连续等技术难点问题仍然亟待解决。论文深入研究了立体匹配的基础理论及局部立体匹配算法中的一些优秀算法,通过对各种算法得到的视差图进行测试对比,最终选择了一种基于重启动随机游走的立体匹配算法进行深入学习,研究其需要改进的方面,并尝试在PC机和嵌入式ARM平台上对其进行实现。经过分析发现,重启动随机游走算法存在边缘权重仅由单一的像素点间的颜色差异决定的问题,并且没有对邻接矩阵进行均衡化,这两个方面的问题会增加错误匹配的概率,降低算法的匹配精度。因此本文提出了一种改进的重启动随机游走算法。首先对边缘权重进行改进,边缘权重由像素点间的颜色差异和空间距离共同决定,然后采用双随机矩阵技术对邻接矩阵进行均衡化,进一步提高匹配精度,最后,通过测试平台对改进算法进行测试,并与原算法的结果对比,测试结果显示,改进算法的匹配精度比原算法有所提高。本文对提出的改进算法进行硬件实现。首先搭建了双目视觉图像采集平台,通过Matlab编程实现左右摄像头同时采集实景图像,并利用Matlab标定工具箱完成了实景图像的立体标定,获得了双目摄像头的内外参数,然后通过OpenCV对采集到的实景图像对进行畸变校正,再用改进算法进行立体匹配,得到实景图像的视差图,完成改进算法在PC机上的实现。而后在虚拟机上搭建Ubuntu系统,再在其之上搭建Linux+QT系统,通过QT调用OpenCV完成图像校正和立体匹配,获得视差图。然后通过交叉编译器编译生成能在ARM开发板上运行的文件,将Linux+QT系统和可运行文件烧写到开发板中,最后通过显示屏显示算法求得的视差图,完成立体匹配算法在嵌入式ARM中的实现。
[Abstract]:With the rapid development of computer vision in recent years, binocular stereo vision, as one of its important research directions, has also been rapidly developed, and its application field has been continuously broadened. The complete binocular vision system consists of image acquisition, camera calibration, image correction, stereo matching and 3D reconstruction. Stereo matching is the most important step. The precision, efficiency and hardware implementation of stereo matching directly determine the performance and practical application of the binocular vision system. Therefore, the study of stereo matching algorithm and its implementation is of great significance and practical value for the development of binocular vision. Up to now, many kinds of stereo matching algorithms have been proposed one after another. However, the technical difficulties in stereo matching, such as weak texture and depth discontinuity, still need to be solved. In this paper, the basic theory of stereo matching and some excellent algorithms of local stereo matching are deeply studied, and the parallax images obtained by various algorithms are tested and compared. Finally, a stereo matching algorithm based on reboot random walk is selected for further study, and its improvement aspects are studied, and it is implemented on PC and embedded ARM platform. After analysis, it is found that the edge weight of the algorithm is determined only by the color difference of a single pixel, and the adjacent matrix is not equalized. These two problems will increase the probability of mismatch. Reduce the matching accuracy of the algorithm. Therefore, an improved random walk algorithm for restarting is proposed in this paper. Firstly, the edge weight is improved. The edge weight is determined by the color difference between pixels and the space distance, then the adjacent matrix is equalized by using double random matrix technology to further improve the matching accuracy. The improved algorithm is tested on the test platform and compared with the original algorithm. The test results show that the matching accuracy of the improved algorithm is higher than that of the original algorithm. In this paper, the improved algorithm is implemented in hardware. Firstly, a binocular visual image acquisition platform is built, and the real scene image is collected simultaneously by the left and right cameras through Matlab programming. The stereo calibration of the real scene image is accomplished by using the Matlab calibration toolbox, and the internal and external parameters of the binocular camera are obtained. Then the real scene image is corrected by OpenCV, then stereo matching is carried out with the improved algorithm, and the parallax image of the real scene image is obtained, and the improved algorithm is implemented on PC. Then the Ubuntu system is built on the virtual machine, and then the Linux QT system is built on the virtual machine. The image correction and stereo matching are completed by QT call OpenCV, and the parallax map is obtained. Then the cross-compiler is used to compile and generate the files that can run on the ARM development board, and then the Linux QT system and the runnable files are burned into the development board. Finally, the parallax map obtained by the display algorithm is displayed. The realization of stereo matching algorithm in embedded ARM is completed.
【学位授予单位】：西南石油大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【相似文献】