【作者】 王金龙；

【导师】 易军凯； 姜大光；

【作者基本信息】 北京化工大学 ， 计算机应用技术， 2008， 硕士

【摘要】 本论文是基于人脸识别与移动检测的视频监控系统而完成的,为了在视频中自动对人脸进行实时的检测、识别,以及对重点监控区域的移动进行检测。本文的主要研究内容为:利用人脸的肤色特征和人眼像素在人脸区域的特殊性和对称性,完成了一个快速人脸检测算法的设计;利用离散余弦变换(DiscreteCosine Transform,DCT)参数进行提取观察向量,并将观察向量用于隐马尔可夫模型(Hidden Markov Model,HMM)模型,完成人脸训练和识别;针对传统移动检测算法对微小移动不敏感的缺陷,提出了一种改进的背景差的算法;将设计好的人脸识别和移动检测算法应用于软件系统当中,能够让用户根据不同的监控条件采用不同的移动检测方法,并设计了录像、警报、记录查看等辅助性的功能模块,方便用户查看以往出现的人脸图像、经过识别后的识别效果以及检测到移动后记录的画面,使系统的功能更加完善;利用设计好的软件系统平台,通过大量的实验,对系统使用的人脸检测和识别算法的效果进行了测试,并将改进背景差算法与传统算法进行了比较,分析它们的适用条件。本系统整体设计结构清晰、功能完善;人脸检测与识别算法效果快速准确,完全适合实时的监控环境;改进背景差算法相对传统的相邻帧差算法更加适合于微小移动的检测,完全可以在设计中将两种方法用于不同的检测场景。更多还原

【Abstract】 This thesis was finished on the basis of a video surveillance system, which includes face recognition and motion detection algorithm and is designed to search and recognize faces appearing in a video sequence and detect motion in selected region.By using skin-color feature, especial location and pixel features of eyes in face area, an efficient face detection algorithm was designed. After face detection, Discrete Cosine Transform (DCT) was used to extract a set of observation, which is provided to train and recognize faces in the way of Hidden Markov Model (HMM). In order to solve the shortcoming that traditional motion detection algorithm can not be used to detect slow moving objects from an image sequence, an improved method was proposed by rebuilding the background. It was coded in the system and users can decide which motion detection method detects slow or fast target. Several modules were added to make the system perfect, including video camera module, alarm module and records reviewing module, etc. On the platform of the completed system, the effects of face detection and recognition have been displayed. Comparison between traditional algorithm of frame difference and improved one of background difference has been tested. The result showed advantage of the improved one, which can be applied in detecting slow moving objects.The face detection and recognition algorithm used in this thesis is real-time and precise, can search and recognize faces quickly.The improved method of background difference is obviously better than the traditional one when slow targets are moving. The two methods can be used in different condition.In the summary, the completed system with detailed structure and function is suitable to be used to real-time video surveillance.更多还原