本文为美国乔治亚理工学院（作者：Keitaro Kamiya）的硕士论文，共136页。

本文提出了一个车辆检测跟踪监控系统的框架。给出了一个优化的目标检测模板，并考虑了该方法在车辆计数应用中的可行性和有效性，实现了基于各路段交通状态速度变化的虚假检测过滤操作和闭塞处理，考虑了跟踪子空间异常仿射变换及其高波动平均加速度数据的技术。结果表明，考虑到真实检测率和虚假检测率之间的权衡关系，该方法具有较好的综合性能。过滤操作在去除大多数非车辆一样移动的物体方面取得了显著成功。采用的闭塞处理技术也提高了系统性能，从而避免了原本会丢失的计数。对于所有测试的视频样本，该框架获得了较高的正确计数率（>93%的正确计数率），同时最小化了错误计数率。在未来的研究中，作者建议针对特定的条件集使用更复杂的过滤器，以及实现用于检测不同阻塞情况的识别分类器。

This thesis presents a framework for motorvehicle detection-tracking surveillance systems. Given an optimized objectdetection template, the feasibility and effectiveness of the methodology isconsidered for vehicle counting applications, implementing both a filteringoperation of false detection, based on the speed variability in each segment oftraffic state, and an occlusion handling technique which considers the unusualaffine transformation of tracking subspace, as well as its highly fluctuatingaveraged acceleration data. The result presents the overall performanceconsidering the trade-off relationship between true detection rate and falsedetection rate. The filtering operation achieved significant success inremoving the majority of non-vehicle elements that do not move like a vehicle.The occlusion handling technique employed also improved the systemsperformance, contributing counts that would otherwise be lost. For all videosamples tested, the proposed framework obtained high correct count (>93%correct counting rate) while simultaneously minimizing the false count rate.For future research, the author recommends the use of more sophisticatedfilters for specific sets of conditions as well as the implementation ofdiscriminative classifier for detecting different occlusion cases.

1. 引言

2. 研究目标

3. 文献回顾

4. 本文所提出框架的文献回顾

5. HAAR级联优化

6. 车辆计数方法

7. 测试、结果与讨论

8. 结论与建议

附录A 用于校正方法的代码

附录B 所有样本的逐车道结果

附录C 所有样本的参考速度与最小速度

更多精彩文章请关注公众号：