学术论文

      基于核协同表示的步态识别

      Gait recognition based on kernel collaborative representation

      摘要:
      为了提高跨视角下的步态识别率,分析识别率低的原因,原因之一归咎于通常采用线性降维的方法进一步提取步态特征,而实际上,特征数据之间可能存在着非线性的关系,故采用核主成分分析法对特征数据进行非线性降维处理,设计了一种核协同表示的步态识别方法.该方法获取步态能量图,采用核主成分分析法对步态能量图数据进行非线性降维处理并提取步态特征,并用协同表示的方法进行分类.实验结果表明:在90°视角下,采用多项式核函数的识别效果明显优于采用高斯径向基核函数的识别效果;该方法在跨视角下取得了显著的识别效果,与其他算法相比,识别率提高了10%以上;该方法的识别速度约是协同表示的识别速度的1~2倍.
      Abstract:
      In order to improve gait recognition rate under cross-view, analyzing reason of low recognition rate, one of the reason is that gait characteristics are extracted using liner dimension reduction method.In fact, there may be a non-linear relationship between the characteristic data.Therefore, kernel principal component analysis is used to reduce non-liner dimension for characteristic data, and gait recognition based on kernel collaborative representation is proposed.Firstly, gait energy image is obtained.Secondly, gait energy image data is reduced non-liner dimension using kernel principal component analysis and gait characteristics are extracted.Finally,collaborative representation is used to classify.Experimental results show that firstly compared with using Gaussian RBF kernel function, polynomial kernel function perform the best result under 90°-view;secondly, this method has achieved remarkable recognition performance under cross-view.Compared with other algorithms, the recognition rate increases by more than 10%;thirdly, the recognition speed of this method is about 1~2 times of the collaborative representation.
      Author: LI Zhan-li SUN Zhuo CUI Lei-lei YUAN Peng-rui
      作者单位: 西安科技大学计算机科学与技术学院,陕西西安,710054
      年,卷(期): 2017, 42(2)
      分类号: TP391.4
      在线出版日期: 2017年5月16日
      基金项目: 中国博士后科学基金资助项目,陕西省教育厅专项科研计划项目