摘要
提出了基于Hu矩的步态周期检测算法,该算法具有尺度、平移不变性,在预处理的标准中心化之前进行,缩短了步态识别前期处理工作的时间,为实时的步态识别提供可能。在分析步态的投影特征具有身份判别的能力之后,进而引出并说明使用Trace变换特征对步态表达的想法是合理的。提出基于Trace变换的步态识别算法,详细地讨论了三种Trace变换的泛函形式,在CASIA(B)步态库上进行验证实验,最佳识别率可达84.14%。这种方法避免了动态时间规整以及线性时间归一等算法的复杂的调整过程。
A novel gait period detection algorithm based on Hu moments was proposed.With the scaling and shift invariance attributes,this method can potentially be used preceding the standardized and centralized image processing.Therefore,the pre-processing time in gait recognition is reduced significantly,which results in the possibility of real-time gait recognition.After analysis of the ability of the projection features of the gait to distinguish identity,it was elicited that it is reasonable of using Trace transform to describe the characteristics of the gait.Following that,gait recognition based on Trace transform was proposed.Three different kinds of functional forms in Trace transform were discussed in detail and experiments were conducted on CASIA-B gait database for validation.The best recognition rate achieved was 84.14%.This method avoids the complicated rectification process of the dynamic time warping algorithm and linear time normalization algorithm.
出处
《吉林大学学报（工学版）》
EI
CAS
CSCD
北大核心
2012年第1期156-160,共5页
Journal of Jilin University:Engineering and Technology Edition
基金
国家自然科学基金项目(60975042
61105057)
中国博士后科学基金项目(20110491087)
盐城工学院人才引进专项基金项目(XKR2011019)
