基于在线增量学习的实时人脸跟踪算法  

REAL-TIME FACE TRACKING ALGORITHM BASED ON ONLINE INCREMENTAL LEARNING

作　　者：包芳[1,2] 张炎凯 王士同 

机构地区：[1]江阴职业技术学院,江苏江阴214400 [2]江苏省信息融合软件工程技术研究开发中心,江苏江阴214400 [3]江南大学,江苏无锡214000

出　　处：《计算机应用与软件》2016年第5期270-273,297,共5页Computer Applications and Software

基　　金：江苏省自然科学基金项目(BK2012128)

摘　　要：提出基于在线增量式极端随机森林分类器的实时人脸跟踪算法。算法用在线极端随机森林分类器实现基于检测的跟踪,并结合动态目标框架和P-N学习矫正检测的错误。实验结果表明,该算法能够在不确定背景下对任意人脸实现较长时间段内的稳定快速的实时跟踪,并能有效排除背景等的干扰,效果较好。The paper proposes a real-time face tracking algorithm,which is based on online incremental extremely random forests classifier. The algorithm achieves detection-based real-time tracking using online incremental extremely random forests classifier,and combines dynamic target framework and P-N learning to correct detection errors. Experimental results show,the proposed algorithm can realise fast and stable real-time tracking for any face in a longer period under uncertain background, and can effectively overcome interferences such as background with preferable effect.

关 键 词：在线增量学习 极端随机森林 P-N学习 动态目标框架 实时人脸跟踪 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]