本文为美国佛罗里达大学（作者：VISHNU KARAKKATNARAYANAN）的硕士论文，共62页。

本文旨在介绍一种新的鲁棒性强的实时系统，该系统能够利用单目视觉快速检测和跟踪视频流中的车辆。该框架是基于Haar特征Viola-Jones分类器的主动学习实现，该分类器集成了Lucas-Kanade光流跟踪器和距离估计算法。首先利用矩形Haar特征建立了一个被动训练的监督系统，一些越来越复杂的弱分类器（本质上是一组退化的决策树分类器）在一开始就被训练。这些弱训练分类器在Adaboost的基础上进行联合级联，形成一个强分类器，该分类器能够消除大部分背景，并在更可能成为候选的图像区域上工作。这将导致速度的提高和虚警率的降低。当模型在独立数据集上评估时，通过查询错误分类的实例，从初始被动分类器生成主动学习模型。该主动训练系统结合Lucas-Kanade光流跟踪器和距离估计器算法，构建了一个完整的多车辆实时检测跟踪系统，然后对建立的模型进行了广泛的静态和真实数据评估，并给出了结果。

This thesis aims to introduce a novel robust real time system capableof rapidly detecting and tracking vehicles in a video stream using a monocularvision system. The framework used for this purpose is an actively learnedimplementation of the Haar-like feature based Viola-Jones classifier integratedwith a Lucas-Kanade Optical Flow Tracker and a distance estimation algorithm. Apassively trained supervised system is initially built by using Rectangular Haar-likefeatures. Several increasingly complex weak classifiers,(which are essentially adegenerative set of Decision Tree classifiers) are trained at the start. Theseweakly trained classifiers are then conjunctively cascaded based on Adaboost toform a strong classifier which results in the elimination of much of thebackground and works on regions of the image which are more likely to be thecandidate. This leads to an increase in speed and reduction in false alarmrates. An actively learned model is then generated from the initial passiveclassifier by querying misclassified instances when the model is evaluated onan independent dataset. This actively trained system is then integrated with aLucas-Kanade optical flow tracker and distance estimator algorithm to build acomplete multi-vehicle detection and tracking system capable of performing inreal time. The built model is then evaluated extensively on static as well asreal world data and results are presented.

1 引言

2 相关研究

3 主动训练的Viola-Jones分类器

4 车辆跟踪与距离估计

5 评估与讨论

6 结论与未来工作展望

更多精彩文章请关注公众号：