复杂动态场景中鲁棒的视觉跟踪算法研究

项目名称： 复杂动态场景中鲁棒的视觉跟踪算法研究

项目编号： No.61201453

项目类型： 青年科学基金项目

立项/批准年度： 2013

项目学科： 电子学与信息系统

项目作者： 温静

作者单位： 山西大学

项目金额： 24万元

中文摘要： 视觉跟踪是实现智能的视频内容分析和理解的重要技术和关键环节。在实际应用中，受到视频中多种因素的影响，视觉跟踪还面临很多挑战。针对背景场景复杂，目标运动存在多样性等问题，本项目拟开展复杂动态场景中鲁棒的视觉跟踪算法研究。本项目将实现视频序列中的无监督背景模型的构建；并结合视觉注意机制获得显著图，在此基础上利用模糊理论对目标进行提取；然后，基于目标提取的结果，通过半监督学习的方法构建目标的外观模型，设计增量式学习的鲁棒视觉跟踪算法,实现对复杂动态场景中长时可靠的跟踪。该研究成果将为运动目标识别和智能视频监控提供有效的工具，并为保障社会安全提供新的技术途径。

中文关键词： 视觉注意机制；显著性检测；目标分割；目标跟踪；

英文摘要： Visual tracking is one of most important techniques for the content based video analysis and understanding. However, it is challenged by many factors, such as, illumination variant, appearance change, and occlusion and so on. In order to perform the long-time ability to track object robustly in complex and dynamic scene, a visual tracking system will be designed in the project. Firstly, an unsupervised learning method will be applied into the background modeling. Secondly, on the basis of the visual attention theory, the object of interest will be extracted by the association of the saliency map and fuzzy theory. Thirdly, a semi-supervised learning method will be adopted for the appearance model, as well as the incremental mechanism for updating the appearance model. Finally, the tracking system based on the combination of the above research results will be explored. The project will be inspirable for the object recognition and intelligent video surveillance, and of great use for the safeguard in the social life.

英文关键词： visual attention scheme；saliency detection；object segmentation；object tracking；