低维度特征的行人检测方法  

Pedestrian detection algorithm using low-dimensional features

作　　者：文韬[1] 李峰[1] 周书仁[1] 

机构地区：[1]长沙理工大学计算机与通信工程学院,湖南长沙410004

出　　处：《计算机工程与设计》2013年第9期3174-3178,共5页Computer Engineering and Design

基　　金：国家自然科学基金项目(60973113);湖南省自然科学基金项目(12JJ6057);长沙市科技计划基金项目(K1203015-11);湖南省标准化战略基金项目(2011031)

摘　　要：针对梯度方向直方图(HOG)算法采用网格密集的大小统一的细胞单元提取行人特征,导致大量高维度的冗余特征问题,提出了低维度特征进行行人检测的算法,建立了以空间金字塔为核心的低维度特征目标模型。该模型通过角点检测算法获取目标轮廓信息,以角点为参考点取16*16像素区域内的梯度方向直方图作为行人特征,利用空间金字塔模型对图像进行分块,按块提取维数统一的特征向量并串联起来形成最终的特征向量。实验结果表明了该方法的准确性和有效性。Aiming at the problems of redundant and high-dimensional features that result from pedestrian features extracted from histograms of oriented gradients （HOG） algorithm by uniform cells, which are intensive, the pedestrian detection algorithm using low-dimensional features is presented, and the target model of low-dimensional features is constructed. Firstly, the contour information of the target is gotten by corner detection algorithm. Then, histograms of oriented gradients （HOG） features are extracted from regions of 16 - 16 pixels in the reference of each corner location. Finally, after dividing the detected images into several sub-blocks by creating the spatial pyramid model, the feature vectors with unified dimension from each block are extrac- ted, and the final feature vectors are generated by concatenating them. The accuracy and validity of the presented method is de- monstrated by the experimental results.

关 键 词：行人检测 角点检测 梯度方向直方图 空间金字塔 低维度特征 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]