本文为德国斯图加特大学（作者：Andreas Harnisch）的毕业论文，共80页。

交通监测与分析是提高道路交通安全的重要内容。通过对交通的监控，在出现危险情况时采取适当的措施，或保持交通流量的稳定。因此，有必要使用一个能够检测和跟踪车辆的系统，以便对交通状况的细节进行评估。区分车辆类型也有助于获得有关交通的更详细信息。在本学位论文中，我们实现并评估了不同的车辆侦测与追踪方法。该实现是在C++中完成程序FloDiEdi的一个插件。实施并评估了三种不同的方法。第一种和第二种方法使用Shi-Tomasi特征检测器和改进的Lucas-Kanade特征跟踪器。第二和第三种方法也使用透视变换。第三种方法采用位置估计和模板匹配相结合的方法对车辆进行跟踪。对于速度计算，第一种方法使用交叉比。第二和第三种方法由于采用了透视变换，使用像素到米的转换来获得以米为单位的距离。提取的信息被可视化地呈现出来，例如ID、车辆类型和速度被显示出来，并另外存储到xml文件中。此外，在xml文件中还存储了车道和车辆大小。摄像系统是一个固定静止的系统，可以捕捉街道的图像。第一种方法的检测率很高，但是跟踪效果不好，所以只有大约10%的车辆被成功跟踪。在第二种方法中，检测率比第一种方法低，但跟踪更可靠，跟踪率约为30%。第三种方法的检测率与第二种方法相同，但成功的车辆数量超过70%。关于计算时间，第三种方法比第一种和第二种方法快7倍，第一种和第二种的计算速度差不多。在对这三种方法进行评价后，三种方法中最快、最准确的是第三种方法，且还有更多的改进空间，例如通过使用GPU进行矩阵运算或并行编程来实现速度提升。

Traffic surveillance and analysis is an important matter toincrease the safety on the roads. By monitoring the traffic appropriatemeasures be taken when dangerous situations appear or to keep the traffic flowsteady. Therefore it is necessary to use a system that can detect and trackvehicles so that details of the traffic situation can be evaluated. Adistinction of the vehicle types is also useful to get more detailedinformation about the traffic. In this diploma thesis different approaches forvehicle detection and tracing via camera systems are implemented and evaluated.The implementation is a plug-in for the program FloDiEdi done in C++. Threedifferent approaches are implemented and evaluated. The first and second approachuse the Shi-Tomasi feature detector and an modified Lucas-Kanade featuretracker. Perspective transformation is used in the second and third approachtoo. In the third approach a combination of position estimation and templatematching is used to track vehicles. For speed calculation the first approachuses cross ratio. The second and third, due to perspective transformation, usea pixel to meter conversion to get the distance in meters. The extracted informationare visually presented such as the ID, the type and the speed of the vehicle isdisplayed and additionally stored to an xml file. Furthermore in the xml filethe lane and the size of the vehicle is stored. The camera system is astationary system which captures the images of the street. The first approachachieves a high detection rate but the tracking does not work well, so justaround 10% of the vehicles were successfully tracked. In the second approach thedetection rate is lower than in the first but tracking works more reliable withan tracking rate of around 30%. The third approach has the same detection rateas the second approach but the amount of successful tracked vehicles is over70%. Concerning the computational time, the third approach is seven times morefaster than the first and second, which are similarly slow. After evaluatingthe three methods the fastest of the three approaches and most accurate approachis the third one. There is still room for more improvements such as achieving aspeed up by using the GPU for matrix operations or parallel programming.

1. 引言

2. 基础知识

3. 从图像中提取知识的层次模型

4. 车辆检测与跟踪方法

5. 车辆检测与跟踪方法

6. 结论与未来工作展望

附录A 车辆检测插件使用说明

更多精彩文章请关注公众号：