智能视频分析系统运动目标检测与跟踪研究

项目名称： 智能视频分析系统运动目标检测与跟踪研究

项目编号： No.61271333

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 无线电电子学、电信技术

项目作者： 谢乾

作者单位： 南京大学

项目金额： 88万元

中文摘要： 在安全防范要求较高的领域，要能及时发现非法行为、可疑目标和及时避免灾害事故的发生，目前智能视频监控能起到一定的预先报警功能，但是漏检、误警率高，无法达到完全满足安防要求。特别是在复杂动态场景中的云、烟、随风摆动的树叶等噪声/干扰给运动目标检测与跟踪的研究带来了极大的挑战，有许多理论和实际的技术难点需要解决。智能视频监控核心部分为运动目标检测与跟踪，它们是实现智能视频监控应用的基础，本课题拟对复杂场景智能视频分析中运动目标检测与跟踪两方面进行深入的研究，主要内容为三个方面： （1）深入研究共生关系的动态场景建模与运动目标检测方法，提取可靠的背景，分离出前景。 （2）基于背景差驱动种子选择的动态场景中自动运动目标的精细分割方法研究。 （3）基于分块表征模型和局部背景估计的自适应视觉目标跟踪算法研究。

中文关键词： 智能视频分析；算法；分割；复杂场景；

英文摘要： In the security precaution fields, monitors need find illegal behaviors 、suspicious targets and potential disaster in the scenes on time . Currently, Intelligent video surveillance technique can alarm early in some special fields , but it still can't satisfy Security requirement because of loss of alarm or mistake warning in complex dynamic scenes. When the dynamic scenes include bothering objects, such as cloud ,smoke ,trees waving , it takes great challenging to study moving object detection and tracking, there are a great number of theory and technology need to be solved . Moving object detection and tracking is key parts to intelligent video surveillance , it also is fundamental to application of intelligent video surveillance , this issue will focus on studying moving object detection and tracking ,it includes three subjects: Firstly, we further study how to modeling the co-occurrence statistics at neighboring pixels and how to detecting moving object in dynamic scenes. Secondly, we will study how to select driven seeds based on a background subtraction and how to segment moving object . Thirdly, we will propose a robust visual tracking algorithm via a patch-based adaptive appearance model driven by local background estimation.

英文关键词： intelligent video analysis；algorithm；segmentation；complex scene；