离散小波变换Haar-LL的行人检测研究

提出一种基于二维离散Haar小波变换的局部二值模式(LBP)与局部梯度模式(LGP)的特征融合方法。对图像进行二维离散Haar小波变换,得到4个不同频率的子图像,对低频部分子图像提取LBP特征,对3个高频部分子图像提取LGP特征,将3个LGP特征并接融合后与LBP特征串接融合进行行人检测。在Matlab环境下利用支持向量机(SVM)对INRIA数据集进行5组实验,分别将该方法与梯度方向直方图(HOG)、金字塔梯度方向直方图(PHOG)、LBP、LGP进行检测率、检测时间、光照鲁棒性以及噪声鲁棒性对比。综合各项实验数据表明,该方法在光照鲁棒性以及噪声鲁棒性方面都能取得更好的效果。

Research on Pedestrian Detection of Discrete Wavelet Transform Haar-LL

This paper presents a two-dimensional discrete wavelet transform Haar Local Binary Pattern（LBP）with Local Gradient Pattern（LGP）feature fusion method Haar-LL. The image of the two-dimensional discrete wavelet transform Haar,to thereby obtains four different frequency sub-images,and extracts the low frequency part of the LBP feature,three high-frequency sub-images of the LGP feature extraction,and takes the three characteristics of LGP parallel fusion and LBP features for serial fusion. Under the Matlab environment using Support Vector Machine（SVM）on the INRIA data set for five experimental groups INRIA dataset experiments carried out on five groups,respectively,with Histograms of Oriented Gradients（HOG）,Pyramid of Histograms of Orientation Gradients（PHOG）,LBP,LGP detection rate,detection time,light and noise robustness contrast. Comprehensive various experimental data show that the robustness of illumination and noise is better.