随着人们对公共场所安全的重视，越来越多的广场、火车站、飞机场都安装了摄像头。但由于摄像头数量众多，监控人员很难在第一时间发现可疑的行为，导致危害公共安全的事件时有发生。智能监控旨在能够帮助监控人员过滤掉大量的无用信息，并通过算法自动识别一些特定的行为，进而保证社会的安全，近些年逐渐成为研究热点。 本文在分析了智能监控系统的国内外研究热点的基础上，针对其中的运动目标检测、遗留物检测算法进行了重点研究，具体工作如下： 第一，本文首先分析了现有运动目标检测算法其优缺点，确定选用背景差分法作为运动目标检测的基本方法。着重研究了基于高斯混合模型和基于码书模型的运动目标检测算法，发现这两种算法的性能受阴影影响很大。然后分析了基于特征的阴影检测方法，并利用其基本思路，改进了码书模型的前景检测算法：通过构建加权平均背景模型，解决了阴影检测算法中对背景图像要求高的问题，使得前景检测算法对阴影有较强的鲁棒性。 第二，本文分析了现有的遗留物检测相关算法。这些算法大都是以多层背景模型为基础，通过控制不同模型的更新速度，比较模型之间的差异判断遗留物。这类算法不仅检测速度慢，而且当遗留物被拿走后，因为模型无法及时更新，会在很长一段时间内出现误检，需要使用“鬼影”检测补偿。本文针对以上缺点，结合码书模型的特点，提出了一种基于历史像素集稳定度的遗留物检测算法。该方法在运动目标检测的基础上，对不属于背景码书模型的像素点记录其之前若干帧像素的信息，构成历史像素集。通过采用统计当前像素与历史像素集的匹配程度来判决该像素点是否稳定，进而判断属于遗留物。 最后，本文在VS2010开发环境和开源计算机视觉库(OpenCV)的基础上，简单实现了相关算法。与传统算法相比，本文的修正运动目标检测检测算法能够较好的抑制阴影，并且满足实时性要求。与传统遗留物检测算法相比，本文算法检测速度更快；当遗留物被移动后，也能更快的排除“鬼影”的影响。

With the rising concern with the security in public places, more and more cameras are put in squares，railway stations, airports and subways. But due to the large number of surveillance cameras, monitoring personnel can hardly find suspicious act immediately. This leads to that the act of endangering public security occurs occasionally. Intelligent monitoring aims at helping monitoring personnel by wiping out large amounts of useless information and recognizing some given acts automatically by algorithms and sustains security of society. It has become research focus in recent years. On the basis of analyzing domestic and foreign hot area of intelligent monitoring, this paper mainly study on moving targets detection and abandoned objects detection. Our analytic works in concrete are mainly as follows: Firstly of all, this paper analyses the characteristic of various kinds of moving target detection and chooses to use background subtraction method to detect moving target in the first place. Especially, the paper studies the Mix of Gaussians Model and Code Book Model and finds they are severely affected by shadow. Then this paper studies shadow detection method based on feature, and uses its idea to improve foreground detection of Code Book Model: by constructing weighted average background model, this paper solves the problem that shadow detection algorithm relies heavily on background image and makes foreground detection algorithm has strong robustness on shadow. Secondly，this paper analyses the existing abandoned objects detection algorithm. Most of them are based on multiple layers of background. Multiple layer background method extracts abandoned objects by controlling the update rate of different layers and comparing the deference of each layer. These methods have slow detection speed. What's more, because the background model cannot update in time, if someone carries the abandoned objects away, these method will lead to error detection. Therefore, these methods need other algorithms to make up this problem. To counter the problems above, this paper utilizes the feature of Code Book Model and presents an abandoned objects detection algorithm based on history pixel set stability. This algorithm is on the basis of moving target detection. It records information of pixels which did not match the background model in the past frames, and uses them to construct a history pixel set. Then it calculates the proportion of a current pixel match the history pixel set and uses the proportion to test whether the pixel stands for abandoned objects. Lastly, the mentioned algorithms are implemented simply by C++ using Open Source Computer Vision Library (OpenCV) in the VS2010 development environment. Experimental results suggest that compared with tradition algorithms, the modified moving target detection algorithm can reduce the influence of shadows and meet requirement of real time. The proposed algorithm is fast in both abandoned objects detection and restraining “ghost” effect.