Gabor小波与HOG特征融合的行人识别算法

A Novel Algorithm for Pedestrian Recognition Based on Gabor Wavelet and HOG Feature

针对传统HOG特征行人检测方法中,当目标存在遮挡以及面对复杂环境条件下,行人识别存在较高漏检率和误检率的问题,提出一种基于Gabor小波与HOG特征融合(G-HOG)的行人识别算法。利用Gabor小波对样本图像进行特征变换并在尺度和方向上融合,获取Gabor特征图像,利用HOG算子在特征图像上提取目标特征,进行样本分类,获取行人目标的疑似区域;对行人样本进行HOG特征提取与训练,实现对疑似区域的目标识别。实验结果表明,基于G-HOG特征的行人识别算法在INRIA、MIT与Daimler数据库上性能表现良好,能够获取较高的查全率和识别率。

Conventional algorithms for pedestrian detection based on histogram of Oriented Gradient（ HOG） feature present weak performance in recognition rate and accuracy when the scenes have complex conditions.In this paper,a novel algorithm is presented for pedestrian recognition based on Gabor wavelet and HOG feature（ G-HOG）.In the first stage,a Gabor wavelet is first applied to obtain Gabor feature.It involves the feature transform and fusion of Gabor feature in the scale and direction. Then,the G-HOG features are generated from the Gabor image.They are finally fed into a classifier to acquire the candidate sample for pedestrian.In the second stage,the HOG features are extracted from the candidate targets. These are then used to recognize the pedestrian based on a two-category discriminator.Because of applying the fusion of Gabor feature and HOG feature,the presented algorithm enhances the recall rate and precision for pedestrian detection.The test results based on INRIA,MIT and Daimler data demonstrate that the presented algorithm has better performance.