【作者】 杨清夙；

【导师】 游志胜；

【作者基本信息】 四川大学 ， 计算机应用技术， 2004， 硕士

【摘要】 车辆视频检测与跟踪系统的主要目标是在不破坏路面的情况下,获取道路车流量、车长、车速、占道率等交通信息,为道路的宏观管理和公路规划设计提供重要科学依据。同时, 动态场景中运动物体的快速分割、光线变化、多车辆粘连、车辆遮挡的处理等问题也给车辆视频检测与跟踪系统的研究带来了一定的挑战。 为了解决这些问题,本文提出了静止摄像机条件下的分别基于线式车辆视频检测器与基于面式车辆视频检测器的运动目标检测与跟踪算法。本文分五个部分进行介绍:(1)简单介绍了车辆视频检测系统的研究背景及实现车辆视频检测与跟踪系统的各种方法。并介绍了目前普遍采用的车辆视频检测器。(2)第二章提出的车辆检测系统采用线式车辆视频检测器和背景消减算法进行运动车辆的检测,通过用户设置虚拟线圈划定检测区域。为了使该系统用于十字路口,采用设置背景缓冲区的方式进行背景的提取与更新,不将停止在路口的车辆捕获为背景。给出实验结果,并比较其性能。该系统能较准确完成运动车辆的计数与测速功能。(3)针对线式车辆检测器的不足,第三章提出的车辆视频检测与跟踪系统采用面式车辆检测器,无须用户设置检测区域,在图像的全部范围内进行车辆的检测与跟踪。采用了高阶统计量与低阶统计量相结合为度量的背景提取算法。自适应背景更新算法则是对每帧图像与背景相比较,从它们的差值进行背景小<WP=3>步长更新。通过类间方差求得最佳分割阈值,采用竖向膨胀算子解决易出现的车辆断层问题,通过种子填充算法找到连通的车辆区域,完成车辆的检测,给出实验结果。针对第一套系统,该系统取得了更为理想的效果,从实验结果中看到提取了完整与准确的背景,背景更新的效果也非常理想,通过阈值分割与其它的数字图像处理算法,检测的结果大部分是正确的。无法通过上述方法解决车辆遮挡问题,但在一定程度上解决了车辆粘连问题。(4)给出了简单的摄像机校准方法,并提出了基于Kalman滤波的运动车辆跟踪技术。在“匹配——修正——预测”过程中实现车辆的运动跟踪。为减少计算量,认为两个坐标无关,采用两个Kalman滤波分别进行两个方向的跟踪。从实验结果看到取得比较理想的运动目标跟踪效果。(5)提出了基于线式检测器与背景消减算法的车辆视频检测系统的实现及设计过程,对每一个模块的功能进行了设计与分析,并给出系统流程。本系统针对车辆视频检测与跟踪系统的一些问题,提出了一些解决方法,通过实验证明,该方法可以运用于实时环境,背景提取与更新算法具有可行性,检测与跟踪结果也比较理想。更多还原

【Abstract】 Research & Implementation of Video Vehicle Detection and Tracking systemThe video vehicle detection and tracking system is used to get traffic information, such as vehicle flow, vehicle length, vehicle velocity and the roads utilization without destroying the roads. It particularly emphasizes on the management of roads such as the traffic management, road design layout etc. Moreover, there are many problems, such as fast segmentation of moving objects, change of light, vehicle sheltering each other, and vehicle blocking, which make many difficulties to the vehicle video detection and tracking system.This paper provides two vehicle video detection and tracking algorithms based on line-type vehicle video detector and area-type video vehicle detector.The paper includes five parts:The first chapter simply introduces the research background of the system,all kinds of ways to implement the video vehicle detection and tracking system and a few of video vehicle detectors which are used widely today.The first video vehicle detection system introduced by Chapter 2 uses the line-type vehicle detector, background subtraction algorithm. And the detection area which is set by operators to detect vehicles. In our system the background buffer is used to obtain and update the right background in order to use the system at the crossroads. At last, the real results are provided and the performance is analyzed. The system can count moving vehicles and calculate the vehicle velocity correctly.According the shortcomings of the line-type vehicle detector, the video vehicle detection and tracking system introduced by chapter 3 uses area-type video detector which uses the whole area as detection area instead of the area set by operators. The background obtainment algorithm uses high order statistics and low order statistics. Adaptive background update algorithm uses small-step-long <WP=5>value, which is determined by the subtraction of the current and previous picture, to update the background. The best threshold value is gotten through the maximum variance. For the vehicle image edge-broken problem, we use dilation operator. At last, seed filling algorithm finished the vehicle detection. The real results are provided to make a comparison with the first system. The second system can get the whole and accurate background with better performance. When the environment changes, the background updating algorithm is effective. The detected results are almost correct through the series of digital image processes, such as threshold segmentation and seed filling algorithm. But we cannot resolve the vehicle block problem completely.The fourth chapter describes the simple camera adjusting method. Meanwhile, the vehicle tracking algorithm based on Kalman filter is advanced. The moving tracking is achieved by three steps that are the matching step, emendation step, prediction step. In order to reduce the compute complexity, the X and Y coordinates are assumed that there is no relationship between them. So the X and Y directions can be tracked separately using the Kalman filters. The results are proved the efficiency of the method.At last, the design and implementation of the video vehicle detection system based on line-type video detector are introduced.The paper provides many ways to resolve the difficult problems of video vehicle detection and tracking system. The system can be used in real-time environment. Moreover, the background obtainment and renewal algorithms are viable and results are perfect.更多还原