摘要
引入了紧致度和矩形度两种新型的多区域特征用于步态识别,并且对同质心高度和伸长度两种特征进行了融合。用中值滤波器估计出步态序列的背景,用差分法提取每帧图像的运动目标轮廓,并在此基础上提取紧致度等多区域特征。基于DTW分类算法在UCSD数据库和SOTON数据库进行了实验。其结果显示:单特征的中紧致度的识别率较高,但总体来说识别率有限,如果把几种特征融合进行乘性融合就能够达到较高的识别率。对实验结果进行分析,从理论上说明了紧致度和矩形度作为主要步态特征的合理性。
Compactness and rectangularity are introduced as new region features and they are fused with center height and extendable characteristic for gait recognition.Background of sequence images is estimated by median filter.Dynamic object silhouettes are extracted by subtracting background,and extracted region features based on silhouettes.Experiments based on DTW classification algorithm in UCSD database and SOTON database show that compactness gets higher recognition rate than other features,but they are all limited;it will get a much higher recognition rate if they are fused.The experiments theoretically show that compactness and rectangularity are reasonable as gait features.
出处
《计算机工程与应用》
CSCD
北大核心
2011年第7期159-161,192,共4页
Computer Engineering and Applications
基金
国家教育部新世纪人才支持计划No.NCET-08-0036~~
关键词
步态识别
紧致度
矩形度
特征融合
gait recognition
compactness
rectangularity
feature fusion
