Content-based Automated Video Surveillance

Surveillance cameras are cheap and ubiquitous, but the manpower required to supervise them is expensive. Consequently the video from these cameras is usually monitored sparingly or not at all. Surveillance camera is a far more useful tool if instead of passively recording footage, it can detect events requiring attention as they happen, and take actions in real time. This is the goal of automatic video surveillance: to obtain a description of what is happening in a monitored area, and then to take appropriate action based on video footages.Increased communication capabilities and automatic scene understanding allow human operators to simultaneously monitor multiple environments. Due to the amount of data to be processed in new surveillance systems, the human operator must be helped by automatic processing tools in the work of inspecting video sequences. In addition to real-time surveillance of events, there is an evident need for after-the-event analysis of stored video.It is the demand of real-time video surveillance and the need of content-based structural surveillance video that spurs our research on content-based automated video surveillance.In this dissertation, a novel framework for content-based automated video surveillance is presented. The ME model constructs an exponential log-linear function that fuses multiple features in codewords to approximate the posterior probability of each layer. Then, we analyze events happened in the monitored area with a HMM's internal state by minimizing the entropy of the joint distribution. Surveillance video shot detection and indexing capabilities are used for online and offline content based retrieval of scenes to be detected.Specifically, we make the following contributions:We present an improved algorithm for background modeling and subtraction. Sample background values at each pixel are quantized into codebooks, which represent a...