低空影像中道路信息检测研究

发布时间：2018-03-07 20:51

本文选题：微型无人机　切入点：交通事故　出处：《中国矿业大学》2017年硕士论文　论文类型：学位论文

【摘要】：交通事故快速处理是智能交通发展的重要领域之一。随着国民经济的快速发展,机动车保有量的快速增加,交通事故越发频繁。传统的交通事故量测方式主要依靠人工量测,作业效率低,较长时间的量测极可能造成交通拥堵,同时量测结果纯粹靠手工记录,勘测精度不高,也无法对事故场景二次验证。随着各类新型测量仪器以及数码照相机应用于交通事故现场勘测,在一定程度上缓解了现场人员的压力,但是仍然避免不了低效率的作业模式。因此,对交通事故现场勘测新技术的研究具有重要的理论价值和实际意义。针对目前火热的无人机市场,本文对无人机低空影像下的道路信息检测方案进行研究,对影像中的车辆及车道线进行目标识别,同时基于识别结果完成标定和图像重绘工作。(1)提出一种基于彩色特征点的车道线检测算子。在分析SUSAN角点检测原理后,提出一种改进的具有针对性的角点检测算法,可以对具有独立明显特征的物体进行检测,并对检测结果进行针对性的聚类,达到车道线检测的效果。实验结果表明在影像较好的情况下,这是一种精度较高的精细化提取方法。(2)提出一种利用车道线宽度进行标定的自标定方法。这种方法可以在道路存在行车线的情况下对影像进行标定,实现图像尺度与现实尺度的转换,实验结果证明这种标定方法的精度比较好,可以满足后续测量工作的要求。(3)对影像中车辆目标的光学特征进行分析,实现了基于多特征融合的车辆检测方法的研究,在分析了常见的各类小型汽车后,构建了一种方便表达的车辆拓扑模型,并与车辆检测的结果进行匹配,基本实现了影像中车辆的检测及矢量化工作。最后,本文利用软件实现了影像中道路信息重绘的一系列功能,基本实现了基于单幅影像的道路信息检测的目的。
[Abstract]:The rapid handling of traffic accidents is one of the important fields in the development of intelligent transportation. With the rapid development of national economy and the rapid increase of vehicle ownership, traffic accidents become more and more frequent. The operation efficiency is low, the measurement pole for a long time may cause traffic jam, at the same time, the result of measurement is recorded purely by hand, and the precision of survey is not high. With the application of various kinds of new measuring instruments and digital cameras to traffic accident scene investigation, the pressure of personnel on the scene has been alleviated to a certain extent. But it still can't avoid the inefficient operation mode. Therefore, the research on the new technology of traffic accident scene survey has important theoretical value and practical significance. In this paper, the road information detection scheme of UAV under low altitude image is studied, and the vehicle and lane line in the image are identified. At the same time, a lane detection operator based on color feature points is proposed based on calibration and image redrawing based on recognition results. After analyzing the principle of SUSAN corner detection, an improved corner detection algorithm is proposed. It is possible to detect objects with distinct and independent characteristics, and cluster the detection results in order to achieve the effect of lane detection. The experimental results show that the images are better. This is a fine extraction method with high precision. (2) A self-calibration method based on lane width is proposed, which can be used to calibrate the image in the presence of road lane. The experimental results show that the calibration method is accurate and can meet the requirements of subsequent measurement work. (3) the optical characteristics of vehicle targets in the image are analyzed. The research of vehicle detection method based on multi-feature fusion is realized. After analyzing the common small vehicles, a convenient vehicle topology model is constructed and matched with the results of vehicle detection. Finally, this paper realizes a series of functions of road information redrawing in image by software, and basically realizes the purpose of road information detection based on single image.
【学位授予单位】：中国矿业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U495

【参考文献】