Application of robust face recognition in video surveil-
lance systems*
In this paper, we propose a video searching system that utilizes face recognition as searching indexing feature. As theapplications of video cameras have great increase in recent years, face recognition makes a perfect fit for searchingtargeted individuals within the vast amount of video data. However, the performance of such searching depends on thequality of face images recorded in the video signals. Since the surveillance video cameras record videos without fixedpostures for the object, face occlusion is very common in everyday video. The proposed system builds a model for oc-eluded faces using fuzzy principal component analysis (FPCA), and reconstructs the human faces with the availableinformation. Experimental results show that the system has very high efficiency in processing the real life videos, andit is very robust to various kinds of face occlusions. Hence it can relieve people reviewers from the front of the moni-tors and greatly enhances the efficiency as well. The proposed system has been installed and applied in various envi-ronments and has already demonstrated its power by helping solving real cases.
Theoretical advances and technical progress have helpedto boost the development of biometrics in recent yearsBiometric technologies exploit the biological character-istics of humans,such as face,fingerprint,iris,vein,andgait to recognize identities.A lot of research work hasbeen done to enable computers to understand and extractthese biological characteristics.
Biometric technologies are applied mainly for twopurposes,identity verification and searching for targetedpersons within database.The former has been founduseful in passage and access control,attendance check-ing,and payment authorization,while the latter is nowexploited for processing a large amount of data,typicallyvideo files,where manual checking is far too exhausting,slow,very inefficient and inaccurate.With the wide ap-plication of monitoring video cameras,or close circuitTV(CCTV)systems,the video signals increase expo-nentially,which provides more and more applicationscenarios for biometrics as a searching algorithm.
In this paper,we propose a system that exploits bio-metric techniques to process massive video files.In anycase,the recorded video signals need to be checkedthrough for any specific individual person or event,andthe proposed system could automatically carry out thewhole process.Especially,the system can effectivelyprocess the occluded face images,hence demonstratesvery high accuracy.
The wide application of video cameras means a hugeamount of video data are produced everyday.Close cir-cuit TV(CCTV)systems have shown their fastestgrowth in the past two decades,and the number is stillgrowing exponentially.Such a surveillance system mon-hors a fixed area and tries to maintain a seamless recordof video coverage of that area.The usage of such sys-terns provides a big resource for event and individualperson tracking.They have found many useful cases inpractice,especially they provided countless examples tohelp police to solve crime cases.
Automatic algorithms have to be adopted in order tomake efficient use of the video data.Most of the systemsare designed to record the video signals while few haveconsidered how to process these recorded data.One ofthe most concerning challenges they face is the lack ofefficient searching mechanisms.In practice,it is stillmostly up to manpower to perform any individual personretrieving job.This not only means the video reviewershave to sit before the monitors for very long hours,butalso means unreliable and inaccurate results.Bio-metric-based algorithms enable the computers to under-stand the person or event characteristics,so the videofiles can be processed饰software instead of humans.
The face recognition algorithms provide a tireless anduniform searching job,but the performance suffers frompoor face image quality.In many cases,the collectedface images are not standard front face images}405},and itis common that a collected face image is partly occludedby glasses,light beard,or simply distorted by lightingeffects.So the image quality would heavily degrade thesystems accuracy of face recognition}3-8}.Therefore,anideal system should offer a face recognition algorithmthat is robust to such facial image distortions,in orderthat efficient searching through surveillance videos for aspecific person can be achieved.
In content retrieval among a huge amount of videos,only automatic and intelligent techniques are feasiblesolution}90o}.The proposed system exploits deep learningalgorithm to perform partial face modeling in the processand substantially improves the accuracy of face recogni-tion with occluded face images.
The system consists of a series of layers.The lowestlayer takes in the video or image signals,the layers in themiddle level perform the face recognition function,andthe highest application layer serves as an interface tosupport the actual security checking or individualsearching job.
Here'face detection'refers to the process of judgingwhether there are human faces in an image.Face detec-tion is the basic stage of face recognition.If there arefaces been detected,they would be then extracted forfurther processing.The proposed system integrates a setof face features obtained from a large number of partialoccluded face images using deep learning methods.Itcan process the incomplete faces and thus improve theperformance under complex scenarios.
Contours and skin tones are important characteristicsof a face.They are generally stable and enable a face tostand out of most background objects,hence,they can beused as good candidates for fast face detection in colorimages.The essential procedure of the feature detectionmethod is developing the skin color model,and using itto detect the pixels of skin color,then loc
剩余内容已隐藏，支付完成后下载完整资料
英语原文共 6 页，剩余内容已隐藏，支付完成后下载完整资料
资料编号：[273953]，资料为PDF文档或Word文档，PDF文档可免费转换为Word
以上是毕业论文外文翻译，课题毕业论文、任务书、文献综述、开题报告、程序设计、图纸设计等资料可联系客服协助查找。
