基于Gentle AdaBoost算法的牛体检测  

Gentle Ada Boost based cow detection

作　　者：高韬[1,2] 李凤民[1] 杨耀权[1] 王菁 

机构地区：[1]华北电力大学自动化系,河北保定071003 [2]奥克兰理工大学知识工程与发现研究所 [3]石家庄市发展和改革委员会信息中心,石家庄050000

出　　处：《计算机应用研究》2016年第11期3495-3498,3512,共5页Application Research of Computers

基　　金：国家自然科学基金资助项目(51306058;71102174);河北省科技支撑计划重点资助项目(15212204D);中央高校基本科研业务费专项资金青年资助项目(2014QN46);国家留学基金委高级研究学者及访问学者(含博士后)资助项目(201406735004);天津市科技支撑重点计划资助项目(15ZCZDNC00130)

摘　　要：针对目标在形状、外观和光照条件发生较大变化时产生的检测率低的问题,以牛体检测为例提出了基于Gentle AdaBoost算法的牛体检测。利用bag of features(BOF)的思想创建特征词典,然后通过词典对牛体目标进行特征提取,最后通过Gentle AdaBoost算法对训练集的BOF特征向量进行训练分类,获得目标对象和场景的分类模型。实验结果表明,该算法训练的检测器在牛体目标存在光照不均匀、形变时均可实现可靠的检测。To solve the problem of low detection rate when targets occur great changes in shape, appearance and light illumi- nation conditions, taking cow detection for example, this paper proposed a Gentle AdaBoost algorithm based cow body detec- tion method. Firstly, it created feature dictionary by improved BOF. Then the cow target was extracted by the dictionary. Fi- nally, it processed training and classifying for the feature vectors of the data set to obtain the classification model of cow object and scene. Experiments show that the detectors trained by the proposed algorithm can achieve reliable detection results even in non-uniform and deformed situations.

关 键 词：牛体检测 BOF特征向量 特征提取 GENTLE ADABOOST 训练分类器 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]