基于双目视觉的可行道路检测方法研究与实现

发布时间：2018-07-07 19:05

本文选题：智能移动设备 + 双目视觉　；参考：《电子科技大学》2017年硕士论文

【摘要】：近年来针对室外场景的智能移动设备研究愈演愈热,交通安全问题已成为该研究领域的热点,而道路检测技术无疑是其中的一大重点。道路检测的主要目的是确定当前环境的可行区域,提醒移动设备沿着特定的边缘躲避障碍,最终到达指定位置。目前道路检测技术发展的难点在于感知道路中的道路区域、道路线和障碍物,传统方法利用的特征主要包括:颜色、纹理、边缘和路标等信息,然而由于道路环境的复杂多变,传统特征已经很难满足实际项目中的各种需求。针对这一问题,本文提出基于双目视觉的可行道路检测方法,在传统特征的基础上增加了道路视差信息,以此提高检测结果的准确度。本文研究内容主要包括:1.研究了基于图像分割的道路区域检测算法。先将图像进行超像素分割或均匀分块,然后在分割的基础上提取特征,利用双目视觉的优点在传统特征的基础上增加视差信息,以此提高分类器检测精度,最后通过相邻关系矩阵或二位直方图进行后处理,进一步提高检测精度。2.研究了基于道路区域的道路线和障碍物检测算法。利用道路区域提取感兴趣区域,提高直线检测的效率,基于颜色差异、道路边缘和道路线距离对道路线进行分类,该算法可以有效地检测出道路边缘线和斑马线。利用道路边缘信息确定障碍物位置,通过视差值得到障碍物的距离信息,基于视频帧间信息计算出障碍物的相对运动情况。最终将道路线和障碍物的检测结果融合,确定了图像中的可行道路。3.搭建嵌入式开发平台,实现了移动场景下的实时道路检测。基于DM8168开发板搭建双目视觉嵌入式硬件平台,将本文算法移植到嵌入式系统中,经过问题分析与优化,实现了USB双目摄像头输入,DSP芯片算法处理,显示器实时展示处理结果的功能。本文通过建立道路数据库进行测试与分析,实验证明本文算法可以有效地将图像中的可行道路检测出来,并且算法执行效率高,可以满足嵌入式运行平台的实时处理要求。
[Abstract]:In recent years, the research of intelligent mobile devices for outdoor scenes has become hotter and hotter. Traffic safety has become a hot spot in this field, and road detection technology is undoubtedly one of the key points. The main purpose of road detection is to determine the feasible area of the current environment and to remind mobile devices to avoid obstacles along a specific edge and eventually arrive. The difficult point of the development of road detection technology is to know the road areas, road lines and obstacles in the road. The characteristics of the traditional methods mainly include the information of color, texture, edge and road sign. However, because of the complex and changeable road environment, the traditional characteristics have been difficult to meet the needs of the actual projects. This paper proposes a feasible road detection method based on binocular vision, which increases the road parallax information on the basis of traditional features, so as to improve the accuracy of detection results. The main contents of this paper are as follows: 1. the road region detection algorithm based on image segmentation is studied. First, the image is segmented or evenly partitioned. Then the feature is extracted on the basis of the segmentation, and the parallax information is added to the traditional feature based on the advantages of the binocular vision. In order to improve the detection precision of the classifier, the detection precision.2. is further improved by the adjacent relation matrix or two bit histogram after processing, and the road line and obstacle detection based on the road area are examined. The algorithm uses the road area to extract the region of interest and improve the efficiency of the line detection. Based on the color difference, the road edge and route distance are classified. The algorithm can detect the road edge line and the zebra line effectively. The location of the obstacle is determined by the road edge information, and the distance letter is worth to the obstacle by the parallax. Based on the information between video frames, the relative motion of obstacles is calculated. Finally, the detection results of road lines and obstacles are fused, the feasible road.3. in the image is set up to build the embedded development platform, and the real-time road detection in the mobile scene is realized. The embedded hardware platform of the binocular vision based on the DM8168 development board is set up. The algorithm is transplanted into the embedded system. After the problem analysis and optimization, the USB binocular camera input, the DSP chip algorithm processing, the display of the display processing results are displayed in real time. This paper tests and analyzes the road database by establishing the road database. The experiment proves that the algorithm can effectively detect the feasible road in the image. Moreover, the algorithm is efficient and can meet the real-time processing requirements of the embedded platform.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U495;TP391.41

【参考文献】