【信息技术】【2011.12】基于单目摄像机的鲁棒汽车检测跟踪系统研究

本文为澳大利亚西澳大学（作者：Soo Siang Teoh）的博士论文，共171页。

本文研究了基于单目摄像机的车辆检测技术，提出了一种能在视频帧中对车辆运动进行可靠检测和跟踪的新系统。该系统由三个主要模块组成：基于对称的车辆提示目标检测、用于车辆验证的两类支持向量机分类器和基于卡尔曼滤波器的车辆跟踪器。

在目标提示阶段，提出了一种快速检测图像中所有可能车辆的技术。这项技术利用了这样一个事实：大多数车辆的前后视图在水平轴上高度对称。首先，利用多尺度对称搜索窗口提取图像中的对称区域和高对称点，然后对高对称点进行聚类，并利用每个聚类的平均位置来假设潜在车辆的位置。通过研究发现，在缩小后的图像上沿多条扫描线进行稀疏对称搜索，可以在不牺牲检测概率的情况下显著缩短处理时间。

需要通过车辆验证来消除提示阶段发现的错误检测目标。研究了基于模板匹配和图像分类的几种验证技术，并对图像特征和分类器的不同组合进行了性能评价。研究结果表明，基于SVM分类器训练的定向梯度特征直方图（HOG）在合理的处理时间内具有最佳的性能。

系统的最后一个阶段是车辆跟踪。本文提出了一种基于卡尔曼滤波器和可靠点系统的跟踪功能，该功能可以跟踪连续视频帧中检测到的车辆运动和尺寸变化。

该系统由上述三个模块集成实现，从而为单目车辆检测提供了一种新的解决方案。实验结果表明，该系统能在不同的天气条件下，有效地检测公路和复杂城市道路上的多辆汽车。

This dissertation investigates thetechniques for monocular-based vehicle detection. A novel system that canrobustly detect and track the movement of vehicles in the video frames isproposed. The system consists of three major modules: a symmetry based objectdetector for vehicle cueing, a two-class support vector machine (SVM)classifier for vehicle verification and a Kalman filter based vehicle tracker.For the cueing stage, a technique for rapid detection of all possible vehiclesin the image is proposed. The technique exploits the fact that most vehicles’front and rear views are highly symmetrical in the horizontal axis. First, it extractsthe symmetric regions and the high symmetry points in the image using amulti-sized symmetry search window. The high symmetry points are then clusteredand the mean locations of each cluster are used to hypothesize the locations ofpotential vehicles. From the research, it was found that a sparse symmetrysearch along several scan lines on a scaled-down image can significantly reducethe processing time without sacrificing the detection rate. Vehicleverification is needed to eliminate the false detections picked up by thecueing stage. Several verification techniques based on template matching andimage classification were investigated. The performance for differentcombinations of image features and classifiers were also evaluated. Based onthe results, it was found that the Histogram of Oriented Gradient (HOG) featuretrained on the SVM classifier gave the best performance with reasonableprocessing time. The final stage of the system is vehicle tracking. A trackingfunction based on the Kalman filter and a reliability point system is proposedin this research. The function tracks the movement and the changes in size ofthe detected vehicles in consecutive video frames. The proposed system isformed by the integration of the above three modules. The system provides anovel solution to the monocular-based vehicle detection. Experimental resultshave shown that the system can effectively detect multiple vehicles on thehighway and complex urban roads under varying weather conditions.

1 引言
2 基于视觉的车辆检测技术综述
3 一种创新的基于对称性的车辆提示技术
4 基于分类器的车辆验证技术
5 基于卡尔曼滤波器的车辆跟踪
6 系统集成、实验与结果
7 结论与未来工作展望
附录A 第四章实验的特征向量文件格式

下载英文原文地址：

http://page2.dfpan.com/fs/clcj8221e2915666c22/

更多精彩文章请关注微信号：