面向无线视频浏览的增量学习人脸检测算法研究  

An Incremental Learning Face Detection Algorithm Oriented to Wireless Video Browsing

作　　者：曾定衡 钟汇才[1,2] 

机构地区：[1]中国科学院微电子研究所,北京100029 [2]中国科学院物联网研究发展中心,江苏无锡214135

出　　处：《小型微型计算机系统》2014年第6期1353-1357,共5页Journal of Chinese Computer Systems

摘　　要：基于人脸检测的视频浏览技术和系统正逐步被接受和推广使用,本文面向无线视频浏览提出了一种增量学习人脸检测方法.主要工作包括:1)提出了融合Gentle Adaboost算法与增量学习的人脸检测方法,该方法能在离线学习部分典型样本的基础上,继续进行增量学习,从而适应不同的监控视频场景,获得更好的检测效果;2)针对无线视频检索与浏览的应用需求,设计并实现了一种基于人脸检测的无线视频检索与浏览系统,可以对检测出来的人脸建立数据库和视频索引,当有检索和浏览需求的时候根据指定的人脸,快速回溯到原始视频进行查看.实验证明,本文提出的增量学习算法有较好的人脸检测性能,并且可以在无线视频浏览系统中进行有效应用.Video browsing technique and system based on face detection has been accepted and put into promotion, this paper has proposed an incremental learning face detection method for wireless video browsing. Our main contribution includes： 1 ） a face detection method which has merged incremental learning with GentleAdaboost algorithm is proposed. It can incrementally learn online after learning some typical samples offline, and therefore adapt to different surveillance video scenariosas well as achieve the better detection performance. 21 Aiming at applicationrequirements, a wireless video retrieval and browsingsystem based on face detection is designed and realized. It can establish database andvideo indexing for detected faces, and playback the original video according toa specific face when browsing is need. Experiments demonstrate that our proposed incremental learning algorithm has better face detection performance and can beeffectivelyappliedtowireless video browsing system.

关 键 词：人脸检测 增量学习 无线视频 视频浏览 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]