作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程

• 开发研究与工程应用 • 上一篇    下一篇

一种模糊森林学习方法及其行人检测应用

周文谊 1,王吉源 2   

  1. (1.江西环境工程职业学院 通讯与信息学院,江西 赣州 341000; 2.江西理工大学 信息工程学院,江西 赣州 341000)
  • 收稿日期:2016-04-22 出版日期:2017-03-15 发布日期:2017-03-15
  • 作者简介:周文谊(1982—),女,讲师、硕士,主研方向为图像处理、物联网技术;王吉源,讲师、硕士。
  • 基金资助:
    江西省教育厅青年科学基金(GJJ14455)。

A Fuzzy Forest Learning Method and Its Application of Pedestrian Detection

ZHOU Wenyi  1,WANG Jiyuan  2   

  1. (1.Faculty of Communication and Information,Jiangxi Environmental Engineering Vocational College,Ganzhou,Jiangxi 341000,China; 2.School of Information Engineering,Jiangxi University of Science and Technology,Ganzhou,Jiangxi 341000,China)
  • Received:2016-04-22 Online:2017-03-15 Published:2017-03-15

摘要: 针对随机森林学习方法训练数据时存在的过拟合问题,通过改进各决策节点的决策函数设计一种模糊森林学习方法。利用高斯隶属度函数构建决策树上各节点的决策函数,将确定决策路径转换为模糊决策路径。根据样本从根节点到叶节点所经过的所有决策节点的模糊决策值乘积生成模糊路径。结合各模糊路径与相应叶节点预测参数得到预测结果。将模糊森林学习方法应用到行人检测领域,分别对Haar特征和方向梯度直方图特征进行学习与分类。实验结果表明,与经典的Adaboost、支持向量机和随机森林分类器相比,模糊森林方法可有效提高行人检测的识别率。

关键词: 行人检测, 随机森林, 高斯隶属度函数, 模糊决策, 方向梯度直方图

Abstract: To solve the over-fitting problem of Random Forest(RF) learning method for training data,a Fuzzy Forest(FF) learning method is proposed by improving the decision function of every decision node.Firstly,it builds decision function of every decision node on decision tree by using Gauss membership functions to convert the clear decision paths to fuzzy ones.Then,it produces fuzzy path with the product of the fuzzy decision values of all decision nodes which the sample goes through from root node to leaf node.Finally,it computes the prediction result according to each fuzzy path and prediction parameter of corresponding leaf node.The new FF learning method is applied in the field of pedestrian detection for learning and classifying both Haar and Histogram of Oriented Gradient(HOG) features.Experimental results show that FF is better than classical classifiers such as Adaboost,Support Vector Machine(SVM) and RF for improving the recognition rate of pedestrian detection.

Key words: pedestrian detection, Random Forest(RF), Gauss membership function, fuzzy decision, Histogram of Oriented Gradient(HOG)

中图分类号: