车辆视频检测与跟踪系统的算法研究

【作者】 张丽；

【导师】 李志能；

【作者基本信息】 浙江大学 ， 物理电子学， 2003， 硕士

【摘要】 随着计算机硬件技术和计算机视觉技术的发展，基于计算机视觉的交通监控系统成为可能，视频车辆的实时检测和跟踪是智能交通监控系统的核心部分。目前存在的检测和跟踪技术在复杂场景下、大范围、多目标的情况下，运动目标的分割和跟踪的效果不是很理想，需要进一步的改善。本文提出了针对运动车辆进行实时检测和跟踪的算法，适用于大面积、多目标的复杂场景，能排除干扰，简单区分车辆和行人，可应用于高速公路和城市交通中。本文提出了基于HSV空间的自适应背景模板的方法来完成对运动目标的检测。与目前几种有代表性的RGB空间的运动目标检测的方法进行了比较，基于HSV空间的自适应模板是一种比较新颖的方法，由于HSV空间和RGB空间相比，对颜色要敏感的多，而且在高对比度的情况下，表现突出。此外，自适应的背景模板，采用图像上的每个象素点的H、S、V的分布为高斯分布建立的三维模型，该模型与实际情况比较接近；在背景、前景的判断上，充分利用了颜色信息；而且在更新算法上也考虑了学习率的问题。使得系统的背景模型既能够满足背景随时间渐变的统计特性；又能够兼顾系统的噪声以及一些突发的干扰因素。具有创新意义的是，本文的HSV空间的自适应背景模型是建立在阴影检测的基础上，能够去除阴影，准确的检测出运动物体。而现今比较流行的运动目标检测的方法，有不少是忽略阴影检测算法的，即使做了阴影检测，也大多基于RGB空间的。本文提出了基于扩展卡尔曼滤波器的运动跟踪模型，来实现对运动目标的跟踪。我们采用扩展卡尔曼滤波器做运动估计，缩小了特征搜索的范围，提高了算法的效率。在特征提取方面，采用了质心和窗口相结合的方法，兼顾了目标的位置和形状；而且还初步的去除了行人对运动目标跟踪的干扰。在特征匹配时，充分发挥了特征值的作用，提出了相似函数的概念。最后在更新算法中，考虑了大范围、多目标的追踪过程中出现的多种情况，如：新目标的出现、旧目标的消失、暂时静止等等，创新性的将图像分成三个区域--进入区、跟踪区、离丌区，设定了各自的临界值，从而进一步保证了算法的可靠性和高效性。从处理的结果看，本文提出的带阴影检测的HSV空间自适应背景模型和卡尔曼滤波运动目标跟踪模型，易于实现运动物体的分割、及阴影检测，能比较准确的实现大范围、多目标的跟踪，而且数学模型简单，运算速度快，系统具有很强的鲁棒性和实用性，能满足实时行进车辆的检测和追踪的要求。更多还原

【Abstract】 With the development of computer hardware and computer vision technology,a computer vision-based traffic monitoring system has become possible. Vehicle detection and tracking real-time system based on video is the key to traffic monitoring system. Many popular related technology didn’t meet vary requirements. Therefore,an efficient and robust detection and tracking system is needed eagerly.The aim to this thesis is to design video-based vehicle detection and tracking system,which can limit the noise of system and pedestrian factor,and used in large area,multiple objects and complex environment in traffic surveillance.The thesis has implemented adaptive background model to detect moving objects. We propose to operate in the Hue-Saturation-Value (HSV) color space,instead of the traditional RGB space,and show that it provides a better use of the color information,and naturally incorporates gray-level only processing. At each instant,the system constructs three Gauss distribution for a pixel and maintains an updated background model,and a list of occluding regions that can then be tracked. However,problems arise due to shadows. In particular,moving shadows can affect the correct localization,measurements and detection of moving objects. This work aims to present a technique for shadow detection and suppression used in adaptive color background model mentioned as before. The major novelty of the shadow detection technique is the analysis carried out in the HSV color space to improve the accuracy in detecting shadows.Moreover,based on these objects segmented,the paper discusses the tracking model based on recursive extended Kalman filter (EKF). Motion estimation based on EKF accurately estimates the feature’s position so as to decreases the size of the region matching feature. In feature extractor,the paper combines centriod of the moving objects to tracking window to distinguish pedestrian and moving vehicle. Similar functions are applied for matching feature. Finally,the system updates adaptive tracking model by means of all kinds of situation,they are new moving objects,moving objects disappearing and pause. We partitions the image to three plots,intake area,tracking region,exit region.Results from experiments show that the model of HSV adaptive background with shadow detection and extended Kalman filter tracking has segmented moving objects and detected shadow so easy and accurately tracked moving vehicles in large area,multiple objects and complex environments. And the system has flexiblemathematic model and can meet real-time and practicality requirements.更多还原