改进型HLBP纹理特征的行人检测  

An improved HLBP texture feature method for pedestrian detection

作　　者：周书仁[1,2] 王刚[1,2] 徐岳峰[1,2] 

机构地区：[1]长沙理工大学综合交通运输大数据智能处理湖南省重点实验室,湖南长沙410114 [2]长沙理工大学计算机与通信工程学院,湖南长沙410114

出　　处：《计算机工程与科学》2016年第5期960-967,共8页Computer Engineering & Science

基　　金：湖南省教育厅科研项目(13B132);综合交通运输大数据智能处理湖南省重点实验室项目(2015TP1005);湖南省交通运输厅科技进步与创新项目(201334)

摘　　要：在行人检测中,Haar型LBP(HLBP)特征采用局部统计方式,有效地降低了噪声影响,相比LBP特征对图像纹理描述有明显优势。但是,HLBP特征在计算特征值时,中心点没有参与计算,导致其信息没有被利用。针对这一不足,提出了改进型HLBP(IHLBP)特征,该方法令中心点参与到计算工作中,并赋予其最大权值。首先利用二维离散Haar小波变换,对图像做两级分解处理,得到三种不同尺度图像;然后针对上述三种图像分别提取IHLBP特征并做归一化处理,最后串接三组特征得到最终的特征向量。在INRIA Person数据集上,采用SVM进行测试。实验结果表明,该方法能有效地提高行人检测识别率。Compared with the local binary pattern （LBP）, the Haar-like LBP （HLBP） can effectively reduce the noise by using local statistics in pedestrian detection. However, when calculating the feature value in the HLBP texture, since the center point is not involved, its information is lost. In order to solve this problem, we propose an improved HLBP （IHLBP） texture feature method, which includes the central point in calculating the feature value with a maximum weight. We realize the two level de- composition operation via the two-dimensional discrete Haar wavelet transform before the IHLBP feature extraction, and obtain three images with different scales. The IHLBP features of the three images of dif- ferent scales are then extracted and normalized. They are series connected and finally the final feature vectors are obtained. Experiments on the INRIA Person data set using support vector machine （SVM） show that the proposed method can effectively improve the recognition rate of pedestrian detection.

关 键 词：行人检测 图像纹理 IHLBP特征 二维离散Haar小波 支持向量机 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]