基于机器视觉的交通压线判别方法研究

发布时间：2018-09-06 09:00

【摘要】：交通是城市发展的支撑,关系着人们生活的方方面面。交通准则是交通发展的根本,随着科学技术发展,智能交通开始步入人们的生活。在智能交通系统(ITS)发展的基础上,本文对交通场景中的车辆压线违章进行了深入的研究,结合了当前成熟的图像处理和机器学习技术,研究了视频中车辆前景的分离,道路路线的检测,车辆压线的判断以及视频流中车牌的检测技术。本文基于高斯背景建模理论,CodeBook建模方法,Hough变换直线检测,支持向量机等方法,对交通压线检测和检测后车辆信息的获取进行了研究。论文的主要研究内容如下:(1)在图像预处理上,分析了HSV颜色模型;对图像使用均衡化技术,并针对不同图像设置不同的均衡范围;研究了几种滤波去噪技术,本文使用了对交通图像有效的双边滤波;对几种边缘检测技术进行实验,最终选取Canny算子边缘检测算法对图像进行分割。(2)对图像二值化阈值的选择,通过两种方法进行研究,最后使用对图像分块求局部阈值,再进行组合的方法对图像二值化;对于视频车辆前景的分割,分别对混合高斯背景建模和CodeBook算法进行了研究,最后使用CodeBook算法作为本实验的分割方法,并对其结果进行处理,使用膨胀算法加漫水填充技术,获取了车辆有效的前景范围。(3)使用Hough变换对车道线进行检测,并对其结果进行处理,获取了有效的车道线位置;研究了图像的矩特征与四边形轮廓包围技术,以获取车辆轮廓的有效范围;使用车辆特征对车辆压线进行判断,并总结了压线过程。(4)针对广泛的交通视觉范围,提出使用支持向量机加图像HOG特征对图像中车牌区域进行检测,通过大量的车辆图像数据进行实验,使用该算法获取了一个高效的分类模型,实现了广泛视觉区域内的车牌定位搜索。
[Abstract]:Traffic is the support of urban development, which is related to all aspects of people's life. Traffic criterion is the foundation of traffic development. With the development of science and technology, intelligent transportation begins to enter people's life. Based on the development of Intelligent Transportation system (ITS), this paper makes a deep research on the traffic scene vehicle line breaking rules and regulations, combines the current mature image processing and machine learning technology, studies the separation of the vehicle foreground in the video. The detection of road route, the judgment of vehicle pressure line and the detection technology of license plate in video stream. In this paper, based on Gao Si's background modeling theory and CodeBook modeling method, Hough transform line detection and support vector machine (SVM), vehicle information acquisition after traffic line detection and detection is studied. The main contents of this paper are as follows: (1) in image preprocessing, the HSV color model is analyzed; the equalization technology is used for the image, and different equalization range is set for different images; several filtering and denoising techniques are studied. In this paper, the effective bilateral filtering for traffic images is used, several edge detection techniques are experimented, and finally the Canny operator edge detection algorithm is selected to segment the image. (2) the binarization threshold of the image is selected and studied by two methods. Finally, we use the method of image segmentation to obtain local threshold and then combine the image binarization. For the segmentation of video vehicle foreground, we study the hybrid Gao Si background modeling and CodeBook algorithm, respectively. Finally, the CodeBook algorithm is used as the segmentation method of the experiment, and the results are processed. The effective foreground range of the vehicle is obtained by using the expansion algorithm and the overflowing filling technique. (3) the lane line is detected by using the Hough transform. The results are processed to obtain the effective lane position; the moment feature and quadrilateral contour are studied to obtain the effective range of the vehicle contour; the vehicle feature is used to judge the vehicle line. And summarizes the process of line pressing. (4) aiming at a wide range of traffic vision, the support vector machine (SVM) plus image HOG feature is proposed to detect the license plate area in the image, and a large number of vehicle image data are used to carry out the experiment. The algorithm is used to obtain an efficient classification model and to achieve license plate location search in a wide range of visual regions.
【学位授予单位】：西安科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】