【作者】 杨超；

【导师】 姚勇； 曹泉；

【作者基本信息】 哈尔滨工业大学 ， 物理电子， 2008， 硕士

【摘要】 车辆跟踪是智能交通系统(ITS,Intelligent Transportation System)中的重要技术,而其中基于视频检测的车辆轨迹跟踪技术由于信息量大、可用范围广,成为许多国家的研究热点。本课题研究的目的在于针对ITS领域中的关键技术,研究基于视频的运动车辆轨迹获取的相关问题。为实现视频的自动检测交通流和交通事件等提供算法前提。本文详细分析了运动车辆轨迹获取中较为常用的方法,根据实验和分析,得到了适合实际应用的视频车辆跟踪算法,本文主要的研究内容包括以下几个方面:(1)根据算法调试、对比及实际应用的要求,使用C++编程语言设计程序,并在Windows系统上设计算法可视化调试平台,然后将程序移植到Linux系统实现调试应用。(2)使用简单的模板匹配算法和图像的仿射变换算法对抖动的视频进行防抖处理,以输出稳定的图像。(3)采用以高斯混合背景模型理论为基础的运动车辆检测算法,实现视频中运动车辆检测,为算法后续操作提供基础信息,通过程序实现将该方法与其它常用方法进行比较分析。(4)使用以光流法为基础结合图像金字塔操作的特征点跟踪模块,该模块能够得到精确的车辆位置变化信息,设计程序实现该算法模块功能,完成轨迹精确位置的获取。(5)利用粒子滤波算法得到车辆运动趋势信息的预测,将该算法与以图像特征为基础的模板匹配算法进行比较分析,评价算法的可行性,增强算法的鲁棒性。本文中基于混合高斯模型的运动车辆检测算法具有较强的适应性,结合粒子滤波的预测功能在提高整个算法鲁棒性的同时,使用特征的跟踪算法为车辆轨迹增加了更为丰富的信息。由于算法整体性能的提高,使其成为实际检测应用中可靠的依据。更多还原

【Abstract】 As an important part of the ITS, the vehicle tracking and the obtaining of the vehicle trajectory become a research focus in the world. According to the key techniques of ITS, some problems related to the moving vehicle trajectory obtaining are researched.In this thesis, several useful methods for vehicle trajectory obtaining are analyzed. Compared with them, the most powerful algorithm is adopted. Then the program system that is up to par of application is designed based on the algorithm.According to the needs of the algorithm debuging, comparing and application, the program on both windows and linux operating systems based on the information of vehicles is designed with C++ program language.In case of the video’s shaking, the template matching is used to estimate the movement’s matrix. The affine algorithm is used to reshape the image to reduce the shaking.Gaussian model theory is used to detect moving vehicles. Then information of vehicles is prepared for next processes of the algorithm. The effect of mixture Gaussian model theory is also compared with other algorithms.In order to improve the stability of the algorithm, Condensation filter is used to estimate the information of moving vehicles. This information then is provided to improve the trajectory matching and updating.Both optical flow method and the pyramid processing are adopted to track features of vehicles. From this step, accurate positions of moving vehicles are obtained.Using the algorithms such as Gaussian model theory, Condensation filter and features tracking correctly, and combining them properly, creditable trajectories are obtained and used in practical surveillance system.更多还原