采用HOG特征和机器学习的行人检测方法  

Pedestrian Detection Using HOG Feature and Machine Learning

作　　者：陈丽枫 王佳斌[1,2] 郑力新 CHEN Lifeng;WANG Jiabin;ZHENG Lixin(Engineering Institute,Huaqiao University,Quanzhou 362021,China;Industrial Intelligent Technology and System Fujian University Engineering Research Center, Huaqiao University,Quanzhou 362021,China)

机构地区：[1]华侨大学工学院,福建泉州362021 [2]华侨大学工业智能化技术与系统福建省高校工程研究中心,福建泉州362021

出　　处：《华侨大学学报（自然科学版）》2018年第5期768-773,共6页Journal of Huaqiao University(Natural Science)

基　　金：国家自然科学基金青年科学基金资助项目(61505059);华侨大学研究生科研创新能力培育计划资助项目(1400222001)

摘　　要：针对基于方向梯度直方图(HOG)/线性支持向量机(SVM)算法的行人检测方法中存在检测速度慢的问题,提出一种将HOG特征与Adaboost-BP模型相结合的行人检测方法.利用边缘检测技术快速检测出行人候选区域,提取出多尺度多方向的HOG特征,利用Adaboost算法训练多个反向传播神经网络用于构建强分类器,实现对测试样本图像的检测识别.结果表明:文中方法具有更高的检测率、更低的误报率和漏检率,具有较好的检测效果.The pedestrian detection method based on the histogram of oriented gradient(HOG)feature and linear support vector machines(SVM)exists a problem with low detection speed.To solve this problem,a pedestrian detection method based on HOG feature and Adaboost BP model is proposed.Edge detection technology is used to detect pedestrian candidate region rapidly and get multi scale and multi direction HOG feature.Adaboost algorithm is used to train multiple back propagation neural network to build strong classifier to realize the detection and recognition of test sample image.Experimental results show that the proposed method has higher detection rate,lower false positive rate and false negative rate has better detection effect.

关 键 词：行人检测 行人候选区域 梯度方向直方图 反向传播神经网络 ADABOOST算法 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]