本文为美国普渡大学（作者：Ka Ki Ng）的博士论文，共134页。

由于视觉监视在安全、执法和军事应用中的重要性，近年来一直是一个非常活跃的研究课题。越来越多的监控摄像头安装在银行、火车站、高速公路和边境等安全敏感区域。由于涉及的数据量巨大，同时由于单调和疲劳，无法保证操作员长时间的警惕性监控。因此，在发生可疑活动时，通常会将视频源存档以用于案件侦破等目的。为了帮助人类操作员识别视频中的重要事件，可以使用“智能”视觉监控系统。这样的系统需要快速而健壮的方法来检测、跟踪和分析运动目标。

在这篇论文中，我们研究了运动目标检测、跟踪和事件分析的方法。我们认为稳健性和计算成本是我们工作的主要设计目标。我们提出的方法能够在室内外环境中、在光照变化和背景动态的情况下检测运动目标。我们还提出了一个快速实现的方法，使用了积分图像的扩展。针对粒子滤波的似然模型，提出了一种动态参数设置的鲁棒跟踪方法。此外，本文还提出了一种基于粒子滤波框架的快速目标跟踪外观模型的构建方法。我们还提出了一种行人“流量”估计方法，该方法计算指定一段时间内通过检测线（跳闸线）或指定区域的人数。该方法基于trip-wire中累积的前景像素数和trip-wire周围区域的纹理特征，对不同的行人流量（拥挤度）检测具有鲁棒性。

Visual surveillance has been a very active research topic in the last few years due to its growing importance in security, law enforcement, and military applications. More and more surveillance cameras are installed in security sensitive areas such as banks, train stations, highways, and borders. The massive amount of data involved makes it infeasible to guarantee vigilant monitoring by human operators for long periods of time due to monotony and fatigue. As a result, video feeds are usually archived for forensic purposes in the event suspicious activities take place. In order to assist human operators with identification of important events in videos an “intelligent” visual surveillance system can be used. Such a system requires fast and robust methods for moving object detection, tracking, and event analysis. In this thesis, we investigate methods for moving object detection, tracking, and event analysis. We consider robustness and computational cost as the major design goals of our work. Our proposed method detects moving objects in indoor and outdoor environments under changing illumination conditions and in the presence of background dynamics. We also present a fast implementation of the method using an extension of integral images. We propose a method for robust tracking with dynamic parameter setting for the likelihood model of particle filtering. In addition, we propose a fast method to construct an appearance model for object tracking using a particle filtering framework. We also present a method for pedestrian “flow” estimation that counts the number of persons passing a detection line (trip wire) or a designated region over a period of time. The method is based on accumulated foreground pixel count in the trip wire and texture features in an area enclosing the trip wire. The

method is designed to be robust to varying pedestrian flow rate (crowdedness).

1.       引言

2. 视觉监控

3. 本文提出的方法

4. 实验结果

5. 结论与未来展望

更多精彩文章请关注公众号：