基于支持向量机的角点检测方法研究  

ON CORNER DETECTION METHOD BASED ON SVM

作　　者：夏辉[1] 穆希辉[1] 马振书[1] 杜峰坡[1] 兰箭[1] 

机构地区：[1]军械工程学院军械技术研究所,河北石家庄050005

出　　处：《计算机应用与软件》2009年第1期230-231,276,共3页Computer Applications and Software

摘　　要：支持向量机是一种基于VC维理论和结构风险最小化原则的统计学习方法,具有小样本学习、推广能力强、解为全局最优等优点。基于支持向量机的角点检测方法,不需要梯度信息,只需对角点样本进行学习,找出角点的特征信息,便可实现对角点的正确检测。通过恰当选取核函数、参数值以及训练样本集,实现了对摄像机标定模板角点的正确检测,并且对畸变角点和含噪声角点也有很好的检测效果。与传统方法相比,所提出的方法具有更强的角点检测能力。Support vector machine （SVM） is a kind of statistical learning method based on Vapnik-Chervonenkis dimension theory and structural risk minimization principle. It has many advantages such as small samples learning, strong promotion ability and global optimal solution. SVM-based comer detection method doesn＇ t need the gradient information, and depending on the feature information that leamed from the comer samples, it can achieve correct comer detection. By properly choosing kernel functions, parameter values and training samples, the method presented in the paper implements correct comer detection for the camera calibration board. At the same time, it also has good ability in detection for aberration comers and noise comers. Compared with the traditional comer detection methods, SVM-based method has stronger ability in comer detectior.

关 键 词：角点检测 标定模板 支持向量机 核函数 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] TP391.41[自动化与计算机技术—控制科学与工程]