【作者】 陈伟强；

【导师】 陈华华；

【作者基本信息】 杭州电子科技大学 ， 信号与信息处理， 2012， 硕士

【摘要】 随着人民生活的提高，我国的汽车保有量正在急速增长。车辆的增加导致了道路交通事件的日益增多，其主要原因之一是车辆的违章行为。有效地预防和确认交通违章行为（如闯红灯、违章变道、违章停车等）并及时阻止进一步发生是摆在我们面前迫在眉睫的任务。随着计算机视觉技术、嵌入式技术、网络通信技术的发展，研究车辆违章行为自动检测系统已经成为当前的一个研究热点。本文搭建了基于IP Camera的车辆违章行为检测的系统，并通过IP网络将带有违章车辆信息的视频图像数据流传输至计算机，从而实现远程监控的功能。具体包括以下几个内容：根据交通道路场景的特点，提出了基于分类模式的背景更新算法，以块为基础，通过块计数触发的方法和一定稳定度的块匹配方法，使得场景有新背景物体产生时保证了背景物体的完整性，并与混合高斯模型法进行了对比，结果表明本文的方法在更新新背景物体时该物体更具完整性。介绍了基于区域的Mean Shift车辆跟踪算法，并在此基础上结合车辆跟踪区域的车辆前景点信息和车辆位置区域信息提出了结合车辆信息的Mean Shift方法，实验表明改进的MeanShift方法取得了更好的跟踪效果。在车辆目标检测和车辆跟踪的基础之上对车辆闯红灯、违章停车、违章压线或变道、违章掉头4种车辆违章行为进行了研究。提出了双虚拟线圈结合车辆跟踪的方法用于闯红灯检测；提出了在基于分类模式背景更新方法下的一定稳定度的块匹配方法，对道路中的违章停车进行检测；提出了道路兴趣区域变换方法，并结合车辆外接矩形框提取和车辆跟踪对车辆违章压线或变道行为进行了检测；利用车辆在道路中行驶时的坐标变化特点以及车辆行驶方向的统计特点实现了车辆的违章掉头检测。以TI的TMS320DM642芯片作为核心处理器，开发了基于IP Camera的车辆违章行为检测系统，并完成了相应的硬件和软件设计，实现了通过普通浏览器查看和保存带有违章车辆信息的交通场景图像的功能。更多还原

【Abstract】 With the improvement of people’s life, the total automobile number in city is growing rapidly.The increase of vehicle on the road leads to the growing of traffic events, one of the main reasons isthe vehicle violation of traffic regulations. To effectively prevent and confirm the traffic violationbehavior (such as red light running, illegal lane change, illegal parking, etc) and then stop its furtherdeterioration is a extremely urgent task. Along with the development of computer vision, embeddedsystem, and network communication, the research of automatic detection system to detect vehicleviolation behavior has been a popular study point.This dissertation builds a detection system, which to detect the vehicle violation of trafficregulations based on IP Camera, to transmit image signal which containing vehicle violationinformation to a computer. This dissertation includes the details as follows.A background updating algorithm based on the classification model, from the characteristics ofroad traffic scene, was proposed, and it ensures the integrity of the object in a scene when a newbackground object forms,by the means of the block counting triggering and the stability of the blockmatching. Experiment shows that compared with the gaussian mixture model method, the proposedis better when updating a new background object.The region-based Mean Shift vehicles tracking algorithm was introduced, and based on it amodified Mean Shift method combined with vehicle information from the vehicle tracking regionalvehicles prospect information and the location was proposed, then the experiment shows theimproved is better.Red lights running, illegal parking, illegal pressure line or lane change, and illegal U-turn werestudied based on the vehicle detection and tracking. Two virtual coils combined with vehicletracking were used to detect red light running, and the certain stability of block matching methodbased on the classification model-based background updating algorithm was used to detect illegalparking, and the interest area transformation combined with external rectangular box vehicleextraction and vehicle tracking was used to detect the illegal pressure line or lane change vehicle,and the changes of coordinates of moving vehicles as well as changes of statistical characteristics ofthe direction of a vehicle were used to detect illegal U-turn.A vechicle violation behavior detection system based on IP Camera with TMS320DM642 wasdeveloped, and corresponding hardware and software were designed. It makes user can easily viewand save the traffic scene image with illegal vehicles information using a ordinary browser.更多还原