基于几何标定和特征检测的隧道图像拼接方法

发布时间：2018-07-05 14:00

本文选题：隧道图像 + 几何标定　；参考：《湖北工业大学》2017年硕士论文

【摘要】：随着我国交通行业迅猛发展,隧道数量越来越多,对隧道进行病害检测是保证隧道安全运营的重要工程,因此隧道病害检测技术也越来越受重视。当前较好的隧道病害检测方法是基于隧道内表面全景图来完成的,做法是在检测车上垂直于行车方向安装一排相机对隧道拍照,需要横向拼接,相机在行车方向不同时刻对隧道拍照需要纵向拼接,再拼接成隧道全景图。但由于隧道自身特点,采集的图像大多鲜有特征、区分度不高,常规的特征匹配方法难以实现拼接。因此,针对隧道图像的拼接方法还值得深入研究。为克服了隧道图像自身特点带来的拼接困难,本文提出了一种隧道图像拼接方法,首先对所有隧道图像都用几何标定和特征检测两种方法拼接。然后对于无特征图像,利用被检测隧道和已标定好的检测系统之间的几何关系,通过惯性导航系统数据减小行车颠簸带来的误差,计算横纵方向相邻图像重合区域,根据相邻图像重合像素点数量叠加而进行拼接,从而完成隧道的几何拼接全景图;对于有特征类图像,采用改进SIFT算法检测特征。隧道检测系统相机辅助光源采用的是高亮度平行光的LED灯,拍摄出的隧道图像中间区域亮,四周区域暗,直接拼接会不利于后期的隧道病害查找与识别。因此,本文将原SIFT算法中图像融合方法加以改进,不同程度地增大拼接后图像在重合区域内的灰度值,能够使拼接区域平滑过渡。最后将特征检测方法拼接的结果覆盖到几何拼接全景图上,替换掉几何标定方法对有特征隧道图像拼接的结果,形成隧道全景图。如此,有隧道病害的图像拼接完全吻合,无隧道病害的图像能等比例还原实际隧道。结合实际算例验证本文提出的方法的实用有效,能达到工程要求,提出的拼接方法为后期隧道病害识别提供支持。
[Abstract]:With the rapid development of traffic industry in China, the number of tunnels is more and more, the tunnel disease detection is an important project to ensure the safe operation of tunnel, so tunnel disease detection technology has been paid more and more attention. At present, the better method of tunnel disease detection is based on the panoramic view of the inner surface of the tunnel. The method is to install a row of cameras perpendicular to the driving direction of the detecting vehicle to take pictures of the tunnel. The camera needs longitudinal stitching to take pictures of the tunnel at different times of driving direction, and then the panorama of the tunnel. However, due to the characteristics of the tunnel itself, most of the collected images have few features, the degree of distinction is not high, the conventional feature matching method is difficult to achieve stitching. Therefore, the method of tunnel image stitching is worthy of further study. In order to overcome the difficulty of tunnel image stitching, a tunnel image mosaic method is proposed in this paper. Firstly, all tunnel images are stitched by geometric calibration and feature detection. Then the geometric relationship between the detected tunnel and the calibrated detection system is used to reduce the error caused by the traffic turbulence through the inertial navigation system data, and the overlap region of the adjacent images in the transverse and longitudinal directions is calculated. According to the number of overlapped pixels of adjacent images, the tunnel geometric mosaic panorama is completed. For the images with features, the improved sift algorithm is used to detect the features. The camera auxiliary light source of tunnel detection system uses LED lamp with high brightness and parallel light. The middle area of tunnel image is bright and the surrounding area is dark. Therefore, the image fusion method in the original sift algorithm is improved to increase the gray value of the stitched image in the same region to some extent, and the smooth transition of the splicing region can be achieved. Finally, the results of feature detection are overlaid on the geometric mosaic panorama, and the geometric calibration method is replaced to form the tunnel panoramic image. In this way, the images with tunnel disease fit perfectly, and the images without tunnel disease can restore the actual tunnel in equal proportion. A practical example is given to verify that the proposed method is practical and effective and can meet the engineering requirements. The proposed splicing method provides support for the identification of tunnel diseases in the later stage.
【学位授予单位】：湖北工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】