智能视频分析系统运动目标检测与跟踪研究

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.

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。

本课题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，消除或降低噪声/干扰对运动目标检测的影响，提取可靠的背景，分离出前景，实现运动目标更精确的跟踪，解决其中的关键技术。.在总结和分析智能视频分析系统目标检测和跟踪研究的研究进展的情况下，完成了复杂场景下动态目标的识别、提取、跟踪算法；一种自上而下的启发式分割方法，用于运动目标的分割；跟踪最新运动目标跟踪研究成果，探索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的理论依据。.开发了嵌入式智能视频分析系统。将复杂场景下动态目标的识别、提取、跟踪算法，通过嵌入式开发，进行软件平台和硬件设备的统一研制。.申请发明专利1项，申请并获得软件著作权3项，形成智能视频分析和目标检测实际软件产品并实现销售。