基于相关性度量的触觉步态特征优化

The haptic force feature optimization based criterion of correlation

在现有动力学特征基础上,提出了基于相关性度量的触觉步态组合特征优化方法(CCFO)。采用数学形态学提取触觉步态区域特征,同时提取触觉步态的图像特征,包括校正后外接矩形长宽比和对比度、相关性、熵等特征;采用相关性度量准则优化得到动力学特征,通过分析图像特征的相关性保留最优图像特征,优化后特征线性叠加构成触觉步态特征集。实验数据采用ITCSH GaitⅡ步态数据库,计算各特征组内相关系数和变异系数,结果表明,各特征具有较好的稳定性,并在身份识别中验证了特征集的有效性,实验结果说明CCFO方法可以有效地减少特征数,提高识别率。

Based on the haptic force dynamics feature,a method of Correlation-basis Combination Feature Optimization（CCFO）is proposed.First,the regional features of haptic force are extracted using mathematical morphology.Meanwhile,image features are extracted,such as the improved ratio of length to width,the contrast,the correlation,and the entropy.Then,the haptic force dynamics feature is optimized by criterion of correlation,and the optimal haptic force image feature is kept through analyzing the correlation coefficient.Finally,the haptic force feature set is obtained by linear summation.The experimental data are taken from the ITCSH Gait II database.The stability of the features is studied by calculating the intraclass correlation coefficient and coefficient of variation.The results show that the CCFO method can effectively reduce the number of features and hence improve the recognition rate in identity recognition.