智能视频监控系统中运动行人分析的研究

发布时间：2018-06-02 19:55

本文选题：视频监控系统 + 运动目标检测　；参考：《南京信息工程大学》2017年硕士论文

【摘要】：随着通信技术与计算机技术的发展,视频监控在日常安防中具有越来越重要的作用。传统的视频监控模式主要是将采集视频数据存储到监控中心的服务器中,仅有单一查看功能,已不能满足用户的需求。这种依靠人工对运动行人分析,不仅缺乏对异常信息及时预警的功能,而且准确率较低。本文基于运动目标检测、行人识别与行人异常行为检测算法展开研究,并设计智能视频监控系统,该系统能够对运动行人异常行为发出警报。首先,为了更好的进行运动目标检测,提出了一种改进的运动目标检测算法。结合图像分块和均值法建立背景模型,运用图像分块与当前帧自适应权重更新背景模型,并采用自适应阈值分割目标,克服了背景中运动残影、光线变化干扰以及前景分割效果差的缺点。实验结果表明,本文方法能够快速准确检测出运动目标。其次,针对单特征行人识别度低的问题,提出了基于多特征融合的行人检测算法。该方法融合了改进的多尺度HOG特征与CSSF特征,全面准确的描述了行人局部特征与全局特征。设计了一种Adaboost强分类器进行行人检测。在INRIA行人库上的实验表明,本文方法大幅提高了行人检测精度。然后,针对行人异常行为检测中存在的问题,在目标跟踪基础上,对行人的形状和运动轨迹特征进行多特征提取,充分描述了行人行为信息,并运用先验知识对上述特征进行量化,检测行人异常行为。实验结果表明,本文方法能够有效检测行人异常行为,包括异常摔倒、异常跑步和徘徊。最后,在上述算法基础上,对智能视频监控系统需求进行分析与设计,利用VisualStudio、OpenCV、Android、javaweb、云服务器等技术进行开发,从而使系统智能化、移动化。实验表明系统运行流程,实时性与准确性较好,能够满足用户远程监控、并及时准确掌握行人异常行为信息的功能需求。
[Abstract]:With the development of communication technology and computer technology, video surveillance plays a more and more important role in daily security. The traditional video surveillance mode is mainly to store the collected video data in the server of the monitoring center, which can not meet the needs of the users only with a single viewing function. This kind of artificial pedestrian analysis not only lacks the function of timely warning of abnormal information, but also has low accuracy. Based on moving target detection, pedestrian recognition and pedestrian abnormal behavior detection algorithms, an intelligent video surveillance system is designed, which can alert the abnormal behavior of moving pedestrians. Firstly, in order to detect moving targets better, an improved moving target detection algorithm is proposed. The background model is established by combining the method of image segmentation and mean, and the background model is updated by image segmentation and current frame adaptive weight, and the target is segmented by adaptive threshold, which overcomes the moving image in the background. The disturbance of light change and the disadvantage of poor foreground segmentation. The experimental results show that the proposed method can detect moving targets quickly and accurately. Secondly, a pedestrian detection algorithm based on multi-feature fusion is proposed to solve the problem of low recognition degree of single feature pedestrian. This method combines the improved multi-scale HOG features with the CSSF features, and describes the local and global features of pedestrians comprehensively and accurately. A Adaboost strong classifier is designed for pedestrian detection. The experimental results on the INRIA pedestrian depot show that the proposed method greatly improves the pedestrian detection accuracy. Then, aiming at the problems in pedestrian abnormal behavior detection, based on the target tracking, the multi-feature extraction of pedestrian shape and trajectory features is carried out, which fully describes the pedestrian behavior information. A priori knowledge is used to quantify the above characteristics to detect the abnormal behavior of pedestrians. The experimental results show that the proposed method can effectively detect abnormal pedestrian behavior, including abnormal fall, abnormal running and wandering. Finally, on the basis of the above algorithms, the requirements of intelligent video surveillance system are analyzed and designed, and developed by using Visual Studio OpenCVN Android Java web, cloud server and so on, so as to make the system intelligent and mobile. The experimental results show that the system has good real-time and accuracy and can meet the functional requirements of remote monitoring and timely and accurate understanding of pedestrian abnormal behavior information.
【学位授予单位】：南京信息工程大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN948.6

【参考文献】