摘要： 由于梯度方向直方图(HOG)特征很难区分与行人具有相似轮廓的物体,并且未能较好利用红外图像中行人轮廓内部的亮度信息。为此,提出一种新的特征——梯度方向和强度直方图(HOGI),将其应用于红外行人检测中。通过支持向量机(SVM)融合多特征的方法,避免多特征串联时维度过高的问题。实验结果表明,与HOG相比,HOGI在不增加特征维度和计算量的情况下,漏报率平均降低50%左右。通过基于滑动窗搜索法对实际红外图像进行检测发现,HOGI+SVM方法比HOG+SVM方法具有更好的检测效果。

关键词: 行人检测, 红外图像, 梯度方向直方图, 强度直方图, 支持向量机, 特征维数

Abstract: Due to the Histograms of Oriented Gradients(HOG) feature is difficult to distinguish the object with similar contour of the pedestrians,and fails to better use the brightness information of pedestrian internal outline in infrared image detection,this paper proposes a Histograms of Oriented Gradients and Intensity(HOGI) feature.The multi-feature fusion method by machine learning can avoid the disadvantage of high dimension while cascading multi-feature.Experimental results show that when the false alarm rate is same,the miss rate of HOGI is reduced by 50% compared with HOG without any increase in feature dimension calculation.Based on sliding window search method,pedestrian detection in real infrared images with the HOGI+SVM has a better performance than HOG+SVM method.

Key words: pedestrian detection, infrared image, Histograms of Oriented Gradients(HOG), histograms of intensity, Support Vector Machine(SVM), feature dimension

中图分类号: 

• TP391