基于挥手行为的性别识别方法

Gender recognition based on hand waving action

提出利用挥手行为进行性别识别的方法.使用基于含时切平面的方法检测周期,用平均剪影表征一个周期序列的挥手行为,PCA降维后利用支持向量机进行分类.实验在60人(30男,30女)的数据库上进行.实验结果表明,用提出的算法从三种挥手行为(挥左手、挥右手、挥双手)中识别出性别的正确率达到89.83%或更高.实验还将人体分成5部分:手臂、头肩、腰、臀和腿,研究人体各组成部分对性别识别的贡献.93个对比实验结果表明,去掉手臂部分识别率下降最快;只通过手臂识别正确率达到86.44%或更高;使用两部分识别,手臂+臀部是最优组合;使用三部分识别,手臂+头肩+腰是最优组合.

A method of gender classification based on human hand waving action was presented.A time-involved-cutting-plane based period detection approach was first applied and then a hand waving sequence of a period was represented by the averaged silhouette（AS）.After dimensionality reduction by PCA,a support vector machine（SVM） was used for the classification.A dataset containing 60 people（30 males and 30 females） was used for the experiments.The experimental results show that applying the proposed algorithm the gender recognition from 3 styles of hand waving,i.e.left hand waving,right hand waving and two hands waving,can achieve the correct rate of 89.83% or higher.A numerical analysis of the contributions of different human components was also presented.The human silhouette was segmented into five components,namely the arm,head,shoulder,waist,buttock,and leg.A number of combinations of the components were then used for gender classification.Ninety-three different experimental results show that： the recognition rate drops most rapidly when the arm component is removed;gender recognition from the arm component can reach the correctness rate of 86.44% or higher;when using two components for gender recognition,the arm-buttock combination works best;when using three components for gender recognition,the arm-head and shoulder-waist combination is the best.