面向机器人导航的立体视觉及目标检测技术研究

发布时间：2018-04-04 21:03

本文选题：摄像机标定　切入点：立体校正　出处：《南京理工大学》2017年硕士论文

【摘要】：本文主要研究了面向智能机器人导航的立体视觉系统,包括摄像机标定、立体校正、立体匹配以及基于立体视觉的道路与车辆检测。本文介绍了当前主流的摄像机标定方法、畸变校正方法以及一种紧凑的立体校正方法,并针对这些方法使用双目摄像机进行了实验。立体匹配是计算机视觉研究领域中一个重要的方面。近年来由全局方法演变而来的局部自适应支持权重算法取得了较好的效果。这些算法的主要问题是,计算量大、复杂性随着窗口大小成比例增加。本文研究了一种以双指数的边沿保持平滑(BEEPS)为基础的立体匹配算法,该方法是非局部的且计算量是线性的,和输入数据、滤波器参数乃至平滑的程度无关。与其他自适应权重算法相比,我们的方法可以得到同等的精度,且更快。本文研究了一种基于立体视觉的道路场景分割及车辆检测的方法。首先,使用动态规划进行立体匹配,并采用左右一致性剔除误匹配点;第二,通过视差图建立V视差,并采用一种基于快速投影的直线检测方法检测直线,并由此计算出消失线以及道路与车辆之间的交界点;第三,建立U视差图检测车辆的水平位置,结合第二步可知车辆位置;最终的实验结果表明本文涉及的方法不仅可以准确的检测出道路,也能够检测出道路上的车辆。
[Abstract]:This paper mainly studies the stereo vision system for intelligent robot navigation, including camera calibration, stereo correction, stereo matching and road and vehicle detection based on stereo vision.In this paper, the main camera calibration methods, distortion correction methods and a compact stereo correction method are introduced, and experiments with binocular cameras are carried out.Stereo matching is an important aspect in the field of computer vision research.In recent years, the local adaptive support weight algorithm evolved from the global method has achieved good results.The main problem of these algorithms is that the complexity increases proportionally with the window size.In this paper, a stereo matching algorithm based on bi-exponential edge preserving smoothing beeps is studied. The algorithm is nonlocal and the computation is linear, independent of input data, filter parameters and even the degree of smoothness.Compared with other adaptive weight algorithms, our method can get the same accuracy and faster.A method of road scene segmentation and vehicle detection based on stereo vision is studied in this paper.First of all, dynamic programming is used for stereo matching, and the left and right consistency is used to eliminate the mismatch points. Secondly, V parallax is established by disparity graph, and a line detection method based on fast projection is used to detect the line.The vanishing line and the junction point between the road and the vehicle are calculated. Thirdly, the U parallax graph is established to detect the horizontal position of the vehicle, and the position of the vehicle is determined by combining the second step.The final experimental results show that the proposed method can not only accurately detect the road, but also detect the vehicles on the road.
【学位授予单位】：南京理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TP242

【相似文献】