【作者】 曾锐利；

【导师】 李刚；

【作者基本信息】 天津大学 ， 精密仪器及机械， 2007， 博士

【摘要】 城市交通监控系统是智能交通系统的重要组成部分之一,传统的交通监控模式已越来越不适应日益增加的城市车辆的需要。本课题将基于电磁感应技术的车辆检测器、射频识别技术以及数字图像处理技术应用于城市交通监控系统中,通过融合处理多种传感器信息,对城市交通实施智能化监控。本文对交通监控中的图像处理、信息融合以及智能控制等关键技术进行了研究。论文的主要研究内容包括:提出了基于帧间颜色梯度的背景建模方法。针对交通视频的特点,将交通画面分区,根据帧间子区域的颜色梯度来检测背景区域,综合子区域背景图像建立背景模型;根据建立的背景模型,采用背景图像差分法提取运动目标,在以自动阈值法进行二值化处理后,利用形态学滤波器进行滤波,并根据颜色特征检测目标的阴影。实验表明该背景建模方法能满足交通监控环境的要求。提出了对目标采取先分类、后跟踪的方法提取车辆的运动轨迹,并判断其违章情况。根据面积和形状复杂度两个特征,以模糊聚类算法对目标进行分类;根据城市交通监控的特点建立约束条件,对分类出的车辆目标,以距离、平均灰度差以及面积差三个特征值建立距离测度函数,通过搜索全局最优匹配的方法来确定相邻帧中车辆目标的对应关系,对车辆目标的轨迹进行跟踪。实验表明,利用这些特征可以对目标进行有效分类和跟踪。提出融合全景和近景摄像机信息进行自动车牌识别。利用全景和近景摄像机同步拍摄交通视频,以近景摄像机的画面识别车牌;利用改进的彩色Sobel梯度算子对图像进行边缘检测,以改进的形态学滤波器处理二值图像;针对车牌候选区域,利用多个特征值作为输入量建立模糊神经网络对车牌进行精确定位;然后对车牌区域进行二值化处理以及矫正,按字符的长宽比分割字符,以模板匹配法识别字符,并对一些相似字符进行多次移位匹配,取最佳匹配结果为识别结果。提出将射频识别技术应用于城市交通监控系统,并对车辆信息以及数据管理方式进行了规划,论述了对多种信息进行融合处理,可以解决一些传统交通管理模式下的难题,在实验中对部分功能进行了仿真测试。采用模糊控制算法,对单十字路口的交通信号优化配时进行仿真分析,提出以空闲时间和车辆延误数来综合评价控制策略,根据评价结果自动调整输入模糊集论域,以适应不同交通状况的配时需要。仿真实验表明调控效果明显。更多还原

【Abstract】 Urban traffic surveillance and control system is one of the most important part in intelligent transportation system, the traditional mode of traffic surveillance and control has been unsuitable to the need of increasing vehicles in urban. In this project, three kinds of techniques, including vehicle detector based on electromagnetism induction, radio frequency identification and digital image processing, are used in urban traffic surveillance and control system. By fusing the information of these sensors, the urban traffic can be monitored and controlled intelligently. In this dissertation, some key techniques, including image processing, information fusion and intelligent control are studied in traffic surveillance and control system.The main content of the dissertation involves:The approach to background modeling based on color grads of different frames is proposed. Aiming at some features of the videos in traffic scene, the image is divided into some sub-regions, every sub-region’s background is established according to the color grads, then the background image is formed by integrating sub-region’s background. According to the background image, the moving targets are extracted by using the background subtraction algorithm, and the image is filtered by morphology filter after processed by dynamic threshold segmentation method, then targets’shadows are detected according as the feature of colors. Experimental results show this method can be suitable to the need of traffic conditions.The improved approach to vehicles’tracks detection is presented, it includes classify targets firstly, tracking targets in succession, then traffic violations are determined. According to two features of area and the degree of shape complex, the targets are classified by using fuzzy clustering algorithm. Some restricted conditions are formed according to some features of urban traffic, and aiming at the vehicles after classification, measure function is established by using distance, average values of gray and difference of areas, then the best matching objects are determined by finding global optimization between two adjacent images, and the vehicles are tracked. Experimental results show targets can be classified and tracked effectively by these features.The method about automatic vehicle license plate recognizition by fusing information from panoramic camera and close-range camera is presented. Videos in an intersection are taken synchronously by panoramic camera and close-range camera, and the images sampled by close-range camera are used to recognize vehicle license plate. Edges in the image are detected by using improved color Sobel gradient operator, then the binary image is processed by improved filters in mathematical morphology. A fuzzy neural network is built by using multi features as inputs, and this network can locate license plate region accurately. The plate region is converted to binary image and rectified, and the characters are segmented according to the ratio of width to height. Then characters are recognized by template matching algorithm, and some similar characters are grouped and matched in shift many times, the results are the matching characters with best matching.The approach to adopting radio frequency identification in urban traffic surveillance and control system is proposed, and vehicle information and data managing mode are planned. In this dissertation, some functions in this system are discussed after fusing multi-information, which are difficult to be achieved in traditional traffic managing mode, and some functions are simulated in experiments.Simulate analysis is carried out to optimize traffic signal timing in single intersection by applying with fuzzy logic controller. A parameter including vacant time and the number of delayed vehicles is presented, the traffic signal control strategy can be reckoned according to this parameter, and on the basis of the reckoning results, the universes of input fuzzy set are adjusted to the need of timing in various traffic conditions. The results in simulation experiments show that this method is effective.更多还原