基于视频检测的车辆变道轨迹识别方法研究

发布时间：2018-03-14 16:19

本文选题：视频检测　切入点：稀疏子空间　出处：《苏州科技大学》2017年硕士论文　论文类型：学位论文

【摘要】：道路交通事故已经成为一个社会公共安全问题,车辆违法变道作为引发交通事故的重要原因之一,越来越受到政府和社会各界的关注。高速公路机动车流量大,且行车速度较高,此时随意变更车道往往会造成极大交通安全隐患,引发交通事故。因此如何有效地检测和识别车辆违法变道轨迹,预防和减少交通事故造成的损失,已成为当今交通管理部门亟待解决的一项紧迫的任务。现阶段我国的车辆违法变道行为识别大都是通过交警现场观测或者对监控视频进行人工判读,这种方法既耗时又耗力并且检测精度较低。针对这些问题,本文提出的基于视频检测的车辆变道轨迹识别方法,可实现无监督状态下车辆随意变更车道行为的识别,该识别模型无需人工判读可直接通过视频检测对车辆变道行为进行判断。为打击不按规定车道行驶、随意变更车道以及高速公路占用应急车道等违法行为提供证据支持。本文主要针对高速公路车辆变道轨迹进行研究,以视频处理技术为核心,利用运动目标检测、车辆识别与跟踪等方法提取视频中运动车辆,获得其运动轨迹并分析运动特征,提出一种基于视频检测的车辆变道轨迹识别方法。主要工作如下:(1)对视频图像预处理的相关算法进行研究,对图像灰度变换、平滑滤波、图像增强以及图像二值化等预处理技术及其相关算法进行比较分析,并通过实验验证,确定本文算法,为后续检测奠定基础。(2)采用背景减差法对运动车辆进行检测,针对高速公路车流量大的特点,选取高斯混合背景建模算法进行背景提取,建立边缘检测算子对图像进行运算以获取运动目标的边缘轮廓以及质心,然后采用暗影提取技术去除图像中的阴影。采用一种改进的稀疏子空间聚类算法对运动目标进行分类,实验证明,该算法能够准确地对场景中的目标进行分类。(3)采用Hough变换检测车道线,目标车辆的跟踪使用改进的形心跟踪算法和模板匹配算法。对运动车辆的轨迹进行提取,并计算已有车辆轨迹与车道线的距离,对车辆变道进行定义,通过车辆轨迹与车道线的距离的方差判断车辆轨迹是否发生违法变道。实验证明,该方法能有效地检测车辆变道轨迹,而且检测的精度和效率较高。
[Abstract]:Road traffic accidents have become a social public safety problem. As one of the important reasons causing traffic accidents, vehicle illegal change has been paid more and more attention by the government and all walks of life. And the speed of driving is high, at this time random change of lane will often cause great hidden danger of traffic safety and cause traffic accident. Therefore, how to detect and identify the track of illegal change of traffic effectively, and to prevent and reduce the loss caused by traffic accident, It has become an urgent task for traffic management departments to solve the problem. At this stage, the identification of vehicle illegal traffic behavior in our country is mostly through traffic police on-site observation or manual interpretation of surveillance video. This method is time-consuming and labor-consuming, and the detection accuracy is low. In view of these problems, this paper proposes a method of vehicle track identification based on video detection, which can realize the identification of vehicle random change of lane behavior under unsupervised state. The recognition model can be used to judge the changing behavior of vehicles directly through video detection without manual interpretation. This paper mainly focuses on the research of freeway vehicle track change, video processing technology as the core, using moving target detection, The methods of vehicle recognition and tracking are used to extract the moving vehicle from the video, obtain the motion track and analyze the motion characteristics. In this paper, a method of vehicle track identification based on video detection is proposed. The main work is as follows: 1) the related algorithms of video image preprocessing are studied, and the gray level transform and smooth filtering of the image are studied. Image enhancement and image binarization and other preprocessing techniques and their related algorithms are compared and analyzed, and through experimental verification, the algorithm of this paper is determined to lay the foundation for subsequent detection. (2) the background subtraction method is used to detect moving vehicles. In view of the characteristics of high traffic, Gao Si hybrid background modeling algorithm is selected to extract the background, and the edge detection operator is established to operate the image to obtain the edge contour and centroid of the moving object. Then the shadow extraction technique is used to remove the shadow in the image, and an improved sparse subspace clustering algorithm is used to classify the moving objects. The algorithm can accurately classify the target in the scene. It uses Hough transform to detect the lane line, and the target vehicle tracking uses the improved centroid tracking algorithm and template matching algorithm to extract the track of moving vehicle. The distance between the vehicle track and the lane line is calculated, and the vehicle path change is defined. The variance of the distance between the vehicle track and the lane line is used to judge whether the vehicle track changes illegally or not. This method can detect the track change of vehicle effectively, and the accuracy and efficiency of the detection are high.
【学位授予单位】：苏州科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U495

【相似文献】