基于视觉的车辆违章轧线与违停监管系统研究

发布时间：2018-07-02 06:49

本文选题：智能交通 + 运动目标提取　；参考：《南京航空航天大学》2014年硕士论文

【摘要】：违章车辆检测与抓拍是智能交通系统中的重要组成部分，在约束司机行为、减少车辆违章与交通事故、提高交通效率等方面发挥着重大作用。本文针对目前车辆越、轧线车道线与违停两种违章行为的监管系统范围小、检测与抓拍成功率较低等问题，研究一种新的基于高速球摄像机的大范围监管系统。首先，根据轧线、违停监管应用需求分析，设计了基于高速球摄像机的系统总体硬件框架、工作原理和算法流程，以增大监管范围和提高可靠性。其次，根据轧线、违停监管应用需求和系统可靠性问题，对现有高速球摄像机缺少角度反馈这一缺陷，将角度编码器增设在其内部云台的两转动轴上，构建了闭环运动控制系统，并研究了基于ARM的测控硬件系统，以此为基础，建立抓拍测控模型。然后，为了提高从道路背景中提取出的车辆目标准确性，分别研究了基于有参与无参背景模型的前景检测算法，通过理论与实验研究了现有算法的优劣，将一种历史像素序列与ViBe相结合，做背景建模，验证实验表明，与现有多种背景建模算法相比，本文算法的目标提取准确率较高，误检率较低，且对初始化时产生的伪前景有良好的抑制能力。此外，针对现有图像测距算法过程复杂、操作繁琐等问题，设计了基于相机成像原理和车道等宽线的标定方法，分别实现了纵向（车道方向）与横向（车道垂直方向）测距，实验结果表明，本文方法实现简单，目标位置信息获取准确，测量误差均在15%以下，较好的达到了本系统动态抓拍要求。最后，针对本系统的大范围监管需要相机配合云台、镜头动作才可以抓拍到不同位置违章车辆高清图像的问题，本文结合抓拍测控模型，设计了动态抓拍方法。为了精确判断目标车辆是否轧线，设计一种根据车辆位置与越线像素比例的轧线判定算法，大大降低非车辆物体引起的误判定，并针对车辆违停设计违停检测算法。整合以上内容，构建车辆轧线与违停监管系统，经多个实际交通路段试点运行，，本系统取得了良好的监管效果。
[Abstract]:The detection and capture of illegal vehicles is an important part of Intelligent Transportation system (its), which plays an important role in restricting drivers' behavior, reducing vehicle violations and traffic accidents, improving traffic efficiency and so on. Aiming at the problems of small scope of supervision system and low success rate of detection and capture of two kinds of illegal behaviors such as vehicle crossing, rolling lane line and illegal parking, this paper studies a new kind of large-scale supervision system based on high-speed ball camera. Firstly, according to the requirement analysis of rolling line and disobeying stop supervision application, the overall hardware frame, working principle and algorithm flow of the system based on high-speed ball camera are designed to increase the scope of supervision and increase the reliability. Secondly, according to the problems of the rolling line, the application requirement of stop control and the reliability of the system, the angle encoder is added to the two rotating shafts of the inside of the cloud head, because the existing high-speed ball camera lacks angle feedback. The closed-loop motion control system is constructed, and the hardware system of measurement and control based on arm is studied. Then, in order to improve the accuracy of the vehicle target extracted from the road background, the foreground detection algorithm based on the participating non-parametric background model is studied, and the advantages and disadvantages of the existing algorithms are studied through theory and experiment. A kind of historical pixel sequence is combined with Vibe to model the background. The experimental results show that compared with the existing background modeling algorithms, the proposed algorithm has higher accuracy and lower false detection rate. And it has a good ability to suppress the pseudo foreground of initialization. In addition, aiming at the complicated process and complicated operation of the existing image ranging algorithms, a calibration method based on camera imaging principle and lane isobath is designed, which realizes longitudinal (lane direction) and lateral (lane vertical) ranging, respectively. The experimental results show that the method is simple, the target position information is accurate and the measurement error is less than 15%, which meets the requirement of dynamic capture of the system. Finally, aiming at the problem that the camera is needed to cooperate with the cloud head and the camera can capture the high-definition image of the vehicle in different positions, this paper designs a dynamic capture method based on the measurement and control model. In order to accurately judge whether the target vehicle is rolling line, a rolling line judging algorithm based on the ratio of vehicle position and crossing pixel is designed, which greatly reduces the misjudgment caused by non-vehicle object, and the algorithm of illegal stop detection is designed for vehicle illegal stop design. By integrating the above contents and constructing the vehicle rolling line and disobeying stop supervision system, the system has achieved a good supervision effect through the pilot operation of many actual traffic sections.
【学位授予单位】：南京航空航天大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：U495

【参考文献】