【作者】 秦武；

【导师】 陈从平；

【作者基本信息】 三峡大学 ， 机械制造及自动化， 2011， 硕士

【摘要】 国家物联网战略的提出给智能交通行业的发展带来了契机。随着传感器技术、网络技术、计算机技术、图像处理技术和机器视觉技术的发展,使现代智能交通系统能够代替人眼实现对交通场景的监管。本文主要研究利用机器视觉技术对高速公路场景图像进行分析和处理,从而实现对运动车辆进行测速(以下简称视觉测速),为智能交通系统(ITS)提供必要的交通监管数据。视觉测速的核心是如何从实时监控图像中准确、快速地提取出指定时间段内车辆实际位移信息。由于同一场景内可能包含多辆运动车辆,因而视觉测速的关键问题转化为如何对运动目标(车辆)进行检测、锁定跟踪、并基于图像定位计算出指定时间段内车辆的实际位移。针对以上问题,本论文的主要工作和取得的成果如下:(一)建立了运动目标的精确检测算法。分析比较了三种常用的运动检测方法,并将基于背景差的目标检测方法作为本文的目标检测方法;应用高斯混合模型对背景进行自适应重建和更新,并提出了合理的改进;开发了基于过渡区提取的目标检测改进算法,能够更准确地获得运动目标的区域;针对高速公路经常出现的雾天情况,开发了适用于高速公路的图像自动去雾算法;(二)改进了适合于高速公路场景的目标跟踪算法。在目标跟踪步骤中针对目标尺度和运动方向的变化分别提出了带宽向下调节策略和目标向前搜索策略,使目标跟踪算法能够对高速公路特定场景的运动目标实现快速自适应跟踪。(三)实现和改进了图像定位算法。研究并实现了基于标定的图像定位方法,同时提出了基于测量的几何定位方法和完全几何定位方法,并提出了一种可行的硬件安装方案,丰富了图像定位理论,使得图像定位方法更全面适应性更强。(四)通过模拟实验和现场实验,对本文提出的目标检测、目标跟踪和图像定位算法进行了验证。根据目标检测与跟踪的结果获得运动目标每一时刻的空间位置,得到目标在监控时段内的运动轨迹,结合图像定位计算得到车辆在公路场景中真实的运动距离,通过帧计时法获得车辆的运动时间,从而计算出车辆在相应时间内的速度。最后提出了一种可行的基于机器视觉的高速公路车辆测速系统设计方案。实验结果表明,本文提出的目标检测方法能够有效提高检测准确度13.9%,测速平均误差控制在2%以内,证明基于机器视觉的高速公路车辆测速理论是可行的,是能够满足实际测速系统要求的。更多还原

【Abstract】 The raise of the Internet of things (IOT) brings the intelligent traffic industry a brighter future prospect. The development of sensor technology, network technology, image processing and machine vision technology makes it possible for modern intelligent transportation systems to substitute human eye in the procedure of traffic monitoring. This paper studies the use of machine vision technology to monitor the scene on the highway, and the scene image analysis and processing speed of moving vehicles to extract the information technology, and attempted to provide the necessary traffic control data for the intelligent transportation systems (ITS).The key point of visually vehicle speed measurement lies in the way of abstracting the practical displacement of vehicles during the specific time scale promptly and accurately. Since there could be numbers of moving vehicles in the same circumstances, the author simplify the problem of the vehicle speed measurement into a key question of the measuring of moving vehicles distance, which include three key steps: vehicle target detection, target tracking and image positioning. Towards this problem the author obtain the conclusion by several steps as follows.(1) Established the accurate detection algorithm of the moving target. The authors compared three commonly used methods of motion detection and object detection based on background difference as a target detection method in this dissertation; application of Gaussian mixture model to adapt reconstruct and update the background and made a reasonable improvement. Taking the fog into account, this paper put forward an automatically Anti-Haze algorithm applicable to the high-way fog scene.(2) The author improved the target tracking algorithm being used in highway scenes. In the steps of tracking the direction of movement for the target scale and bandwidth of proposed changes in regulation strategy and goals down forward search strategy had been proposed, which made the target tracking algorithm adaptive to track fast moving targets in the specific scene on the highway.(3) Realized and improved the calibration orientated image localization method. Therefore proposed location method based on geometric measurements and full geometric positioning method. The author also proposed a hardware installation program with reliable feasibility which enriched the image positioning theory and the all-situation adaptability of the image localization.(4) In this dissertation, through simulation and field test, the proposed target detection, target tracking and image positioning algorithm had been implemented. According to the results of target detection and tracking of moving targets the author obtained the spatial location of each moment and the target trajectory in the monitoring period, combined with image positioning algorithm can the author obtain the real distance that the vehicle went through in the road scene. Calculating the real time of moving away from the scene, obtained the frame timing of the speed of vehicles. Through the frame timing method the author obtained the vehicle’s moving time thus calculate the vehicle speed in the corresponding period. Finally a practical machine vision-based highway vehicle speed system design. Experimental results show that the proposed target detection method can effectively improve the detection accuracy by 13.9%, and the average error can be controlled within 2%, these results proved that machine vision-based highway vehicle velocity measurement theory is feasible, is able to meet the system requirements for the actual velocity measurement.更多还原