摘要
针对步态识别性能易受视角、着装和携带物品等变化影响的问题,提出了一种基于修正步态能量图和视角检测的步态识别方法。首先,对步态能量图进行修正,降低着装和携带物品的变化对步态识别的影响;接着,基于修正的步态能量图提取熵特征,并依据最近邻准则检测步态序列的视角;最后,在相同视角的数据库下,采用二维主成分分析和二维线性判别分析相结合的方法提取步态特征,并采用最近邻准则进行分类,以降低视角变化对步态识别的影响。通过在CASIA B数据集上进行对比实验,证实所提方法对视角、着装和携带物品等变化的鲁棒性强,平均识别率高。
For solving the problem that the variety of view,clothing and carried objects influence the performance of gait recognition, a gait recognition method based on modified gait energy image and view detection was proposed. Firstly, the method modifies the gait energy image, to reduce the impacts from variety of clothing and carried objects. Then, it ex- tracts entropy features from modified gait energy image, and detects the view of unknown gait sequence according to nearest neighbor criterion. Finally, it extracts gait features by combining two-dimensional principal component analysis and two-dimensional linear discriminannt analysis, and executes classification by using nearest neighbor criterion, to re- duce the impacts from view variety. Experiments implemented on CASIA B dataset show that, the new method has a high average recognition rate, and is robust to the variety of view, clothing and carried objects.
作者
李晶
张菁
倪军
LI Jing ZHANG Jing NI Jun(College of Computer Science and Technology, Harbin Engineering University, Harbin 150000, China College of Computer Science and Information Engineering, Harbin Normal University, Harbin 150000,China Carver College of Medicine, The University of Lowa, Lowa 52240, America)
出处
《计算机科学》
CSCD
北大核心
2016年第8期300-303,308,共5页
Computer Science
基金
黑龙江省自然基金项目(F201231)资助
关键词
步态识别
步态能量图
视角
熵
最近邻
主成分分析
线性判别分析
Gait recognition, Gait energy image, View, Entropy, Nearest neighbor, Principal component analysis, Linear discriminate analysis
