【作者】 王军伟；

【导师】 毛恩荣；

【作者基本信息】 中国农业大学 ， 车辆工程， 2003， 博士

【摘要】 利用计算机视觉技术进行交通状况检测与信息采集是智能交通系统(ITS)领域中的一个重要课题，而运动车辆的检测与跟踪则是其中最基础的部分。本文对此问题进行了系统和深入的研究，提出了一种在自然道路背景下运动车辆自动检测与跟踪的方法，并设计开发了相应的软硬件系统，初步实现了运动车辆的自动跟踪。 本文所进行的研究主要包括以下方面：(1) 在图像差分算法基础上，研究提出了运动车辆自动检测的栅格算法，该方法通过计算当前帧与参考帧对应栅格的不相似度来检测是否有运动车辆进入视场；定义了不相似度下降率DSDR，基于此，可以比较准确地确定出运动车辆在栅格中的位置，并可方便地将车辆图像作为模板保存下来。(2) 采用Kalman滤波理论建立了运动车辆状态预测线性模型，该模型采用8自由度的向量来描述系统状态；利用面向对象的思想，以类的形式对预测模型进行了封装实现，并给出了类的数据成员和成员函数。(3) 讨论了在摄像机固定情况下和在摄像机运动情况下运动车辆的跟踪问题。在摄像机固定情况下，以运动车辆的位置估计为中心，按照搜索策略，通过模板匹配对运动车辆进行跟踪。跟踪实验结果表明，提出的搜索策略是有效的。在摄像机运动情况下，建立了运动车辆跟踪坐标系，并得出了运动车辆跟踪坐标系与预测坐标系之间的转换关系；通过摄像机-云台系统所采用的视觉坐标系，推导出了摄像机-云台系统转动的角度控制公式，为运动车辆的跟踪提供了理论基础。(4) 根据面向数据流的设计方法，在综合运动车辆自动检测、状态预测和自动跟踪算法的基础上，采用visual C++作为开发语言，在Windows 2000平台上设计开发了相应的软件系统。软件开发时，以软件工程学模块化思想为指导，将系统划分为图像采集与提取、图像数据处理、图像显示等六个模块。所开发的软件系统能够较好地完成运动车辆的自动检测、状态预测，并初步实现了运动车辆的自动跟踪。(5) 介绍了跟踪系统硬件部分的组成和结构。更多还原

【Abstract】 The application of computer vision technology to traffic detecting and information collecting is an important subject in ITS (Intelligent Transportation System), and moving vehicle detecting and tracking is the basic part. The further study on this question is carried out systematically in this paper. Taking the natural road as the background, a kind of moving vehicle auto-detecting and auto-tracking method is brought forward. Moreover the software system is designed and implemented. A moving vehicle can be automatically tracked elementarily by using the software system.The main contents of the study include such aspects as follows: (1) The moving vehicle auto-detecting grid arithmetic is studied and presented on the basis of the idea of the image difference arithmetic. The moving vehicle is detected whether or not to enter the field of view by calculating the grid dissimilarity between the current frame image and the reference frame image. DSDR (Dissimilarity Descend Rate) is defined, and by calculating it the position of the moving vehicle in grids can be determined exactly, so the vehicle model image can be saved conveniently. (2) The moving vehicle state forecast linear model in which the system state is characterized by an eight-dimension vector is founded by adopting Kalman filter theory. The forecast model is encapsulated as a class with object-orientated technology. The data members and member functions of the class are also given. (3) The moving vehicle tracking problem is discussed at two conditions: the camera is still and the camera is moving. Under the circumstance that the camera is still, the moving vehicle is tracked by model matching according to the searching strategy that is proved valid by tracking experiment. Under the circumstance that the camera is moving, the tracking rectangular coordinates are founded, and the relationship between the tracking rectangular coordinates and the forecast rectangular coordinates is given. The angle control formula of the camera & pan-tilt-device system is educed with its vision coordinates, which provides theory basis for moving vehicle tracking. (4) The data stream chart of the system is given by making use of the data-stream-orientated design method. The software system is implemented with Visual C++ program language on Windows 2000 platform. With the guidance of the modularization idea, the software system is divided into six modules: image collecting and taking module, image processing module, image displaying module, and so on. This software system can accomplish moving vehicle auto-detecting, state forecasting and auto-tracking elementarily. Some useful practice in moving vehicle video surveillance is done. (5) Last, the composing and the structure of the hardware system are introduced.更多还原