【作者】 林婷；

【导师】 卢官明；

【作者基本信息】 南京邮电大学 ， 信号与信息处理， 2012， 硕士

【摘要】 智能视频监控系统能够从摄像头拍摄的视频图像序列中检测、跟踪和识别运动目标，继而分析理解运动目标的行为，从而完成智能安全防范的任务。视频监控中的异常行为检测能够将监控系统中大量对安防无用的信息忽略掉，从而节省大量的人力物力，是近年来被广泛关注的研究领域。它以摄像机拍摄到的视频图像序列为研究对象，以准确性较高的运动目标检测、跟踪和分类为研究基础，以实现目标的行为分析理解的研究目的。本文基于运动目标分析和运动轨迹分析，对视频监控中的边界越线、遗留物以及人体滞留徘徊这些简单异常行为的检测算法进行了一些探索性的研究，本文的主要工作如下：1．分析研究了作为异常分析检测基础的运动目标检测、跟踪和分类的几种主要算法。2．分析了运动目标分析和运动轨迹分析的研究方法。在运动目标分析方面主要研究了目标区域的描述方法、运动目标的表示方法和运动目标的基本特征等方面的内容。在运动轨迹分析方面主要给出了基于距离-方向角的运动轨迹表示方法和运动轨迹曲线主方向角的求解方法。3．采用基于运动空间的方法，结合运动目标分析和运动轨迹分析所提供的运动目标的运动特征信息，给出了边界越线和遗留物异常检测方法，以及人体滞留徘徊异常行为检测的方法。其中，改进了遗留物异常检测中背景更新的算法。最后，本文在VC++6.0开发环境和OpenCV计算机视觉类库相结合下实现了相关算法，并对实验结果进行了分析。实验结果表明该算法能够正确、快速地实现视频图像序列中边界越线、遗留物的异常检测，以及人体滞留徘徊异常行为的检测，同时能够满足系统所需的实时性要求。更多还原

【Abstract】 Intelligent Video Surveillance (IVS) is a kind of security technology to automate objectdetection, tracking, recognition, and activity analysis in image sequences, which are captured bycamera. The analysis and recognition of human abnormal behavior of IVS, which are being moreand more popular, can not only ignore a large of useless information, but also save a lot of humanand material resources. Its research object is the image sequences captured by camera, and it aims atunderstanding the object’s action based on object detection, tracking and recognition.This paper studies simple human abnormal behavior of IVS such as crossing line, abandonedobject and hovering. The main contributions of this thesis are summarized as follow:1. Make a further study on the algorithm of object detection, tracking and recognition.2. Make a further study on the research method of motion analysis and motion trajectoryanalysis. Research on motion object analysis includes the descriptive method of object area andobject, and the motion feature of the object. Research on motion trajectory analysis includes themotion trajectory’s representation based on distance-direction angle, and the solving method ofmain direction angle.3. This paper gives the recognition method of achieving the abnormal detection such ascrossing line and abandoned object, and recognizing the simple human abnormal behavior hovering,relying on activity analysis based on motion space and the motion feature proposed by the motionanalysis and motion trajectory analysis. Furthermore, improve the background update algorithm inhuman abnormal behavior of abandoned object.At last, this paper accomplishes the algorithm in the environment of VC++6.0 and OpenCV,and analysis the experiment results. Experimental results certify that the algorithm can fast andaccurately achieve the abnormal detection such as crossing line and abandoned object, andrecognize the simple human abnormal behavior hovering in image sequences. In addition thealgorithm can meet the needs of real-time of the system.更多还原