Content-based video analysis for visual surveillance and video interpretation

本研究的目的是开发新的方法和算法的视频内容的分析。我们的目标是从实时视频流和视频记录中提取有意义的信息。所提取的信息可以是关于场景内容及其结构或者关于正在发生的动作。特别是，我们将开发的方法中使用的智能视觉监控的情况下，感兴趣的事件必须可靠地检测和分类。我们将研究的问题，可靠的长期跟踪的移动对象，下闭塞，和跨相机。我们的策略将是建立一个环境模型所观察到的场景包含结构信息的静态元素和表示的方式移动对象通常与场景交互。我们还希望开发方法来改进事件和对象的自动分类，部分地使用用户的反馈，以便更准确地识别最相关的事件，从而在检测到此类事件时采取最适当的行动。

The objective of this research is to develop new methods and algorithms for the analysis of video content. Our goal is to extract meaningful information from live video streams and from video recordings. The extracted information could be about the scene content and its structure or about the actions that are occurring. In particular, we will develop approaches to be used in the context of intelligent visual surveillance where events of interest must be reliably detected and classified. We will study the problem of the reliable long-term tracking of moving objects, under occlusions, and across cameras. Our strategy will be to build an environmental model of the observed scene containing structural information about the static elements and representation of the way the moving objects typically interact with the scene. We also want to develop approaches to improve the automatic classification of events and objects, partly using the user's feedback in order to more accurately identified the most relevant events and therefore take the most appropriate action upon detection of such event.