双目立体视觉测距技术研究

发布时间：2018-12-11 11:03

【摘要】：随着工业4.0的提出,拥有感觉要素的智能机器人已成为推动现代工业发展的重要因素。双目立体视觉是实现机器人自动测距的重要技术手段,广泛应用于现代工业中的机器人导航与定位、空间三维立体的跟踪与重构、板料成形的动态测量、极限曲线的检测等领域。近年来,双目立体视觉的研究进展飞速,但仍有问题有待更好的解决,亟需研究人员提出新的解决办法。这些问题主要集中在两个方面:一是如何进一步地提高摄像机标定的精度,二是如何平衡立体匹配的精度与速度。因此,本文对双目立体视觉中的摄像机标定精度与立体匹配展开了重点研究。为了提高摄像机标定精度,本文对亚像素角点检测算法进行了部分改进,在通过Harris角点检测算法得到像素级角点的基础上,利用加权灰度梯度和二项式拟合的方法得到的亚像素级角点的位置,实验结果表明该方法提高了摄像机的标定精度。为了提高双目立体视觉的匹配精度,降低匹配算法在视差不连续区域和噪声干扰下的误匹配率,本文选用了基于Census变换的立体匹配算法,主要工作包括以下三方面:在Census变换阶段,计算变换窗口八邻域像素的平均值代替中心像素作为参考像素进行Census变换,提高单像素匹配代价的可靠性;在代价聚合阶段,应用方框滤波进行代价聚合,提高匹配速度;在视差匹配阶段,通过对粗略视差图的左右一致性检测、不连续性调整、亚像素求精,提高最终视差图的质量,实验结果表明本文改进的算法提高了低纹理及视差不连续区域的视差精度,降低了误匹配率和运算复杂度,提高了双目视觉的稳健性,同时保持了较好的实时性。搭建双目立体视觉测距平台进行测距实验,实验结果表明,本文的双目立体测距平台有较高的测距精度,在实际工程中有较大的应用潜力。
[Abstract]:With the development of industry 4.0, intelligent robot with sensory elements has become an important factor to promote the development of modern industry. Binocular stereo vision is an important technique to realize robot automatic ranging. It is widely used in modern industry, such as robot navigation and positioning, three-dimensional space tracking and reconstruction, dynamic measurement of sheet metal forming. Limit curve detection and other fields. In recent years, the research of binocular stereo vision has made rapid progress, but there are still some problems to be solved better. These problems mainly focus on two aspects: one is how to further improve the accuracy of camera calibration, the other is how to balance the accuracy and speed of stereo matching. Therefore, this paper focuses on the camera calibration accuracy and stereo matching in binocular stereo vision. In order to improve the accuracy of camera calibration, the sub-pixel corner detection algorithm is partially improved in this paper. Based on the Harris corner detection algorithm, the pixel level corner is obtained. The position of sub-pixel corner is obtained by weighted grayscale gradient and binomial fitting. The experimental results show that the calibration accuracy of the camera is improved by this method. In order to improve the matching accuracy of binocular stereo vision and reduce the mismatch rate of the matching algorithm under disparity discontinuous region and noise interference, the stereo matching algorithm based on Census transform is selected in this paper. The main work includes the following three aspects: in the stage of Census transform, the average value of the eight neighborhood pixels in the transform window is calculated to replace the center pixel as the reference pixel to carry out the Census transform to improve the reliability of the matching cost of the single pixel; In the stage of cost aggregation, the block filter is applied to the cost aggregation to improve the matching speed. In the parallax matching stage, the quality of the final parallax map is improved by checking the left and right consistency of the rough parallax map, adjusting the discontinuity, and refining the sub-pixel. The experimental results show that the improved algorithm improves the parallax accuracy of low texture and parallax discontinuous regions, reduces the mismatch rate and computational complexity, improves the robustness of binocular vision, and maintains better real-time performance. The experimental results show that the binocular stereo ranging platform has high ranging accuracy and has great application potential in practical engineering.
【学位授予单位】：西安理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】