结合图模型的优化多类SVM及智能交通应用  

An optimal multi-class SVM combined with graph model and its application on intelligent transportation

作　　者：王剑[1] 张伟华[2] 李跃新[3] 

机构地区：[1]常熟理工学院计算机科学与工程学院,江苏常熟215500 [2]郑州成功财经学院信息工程系,河南郑州450000 [3]湖北大学计算机与信息工程学院,湖北武汉430064

出　　处：《电子技术应用》2017年第2期132-136,共5页Application of Electronic Technique

基　　金：江苏省高校自然科学研究项目(12KJB520001)

摘　　要：为提高多类支持向量机分类器对多目标的分类准确度,提出一种结合无向图模型优化的多类支持向量机分类器。首先,利用余弦测度计算训练数据之间的相似度,构建包含训练数据和相似度矩阵的无向图模型,求解相似度约束矩阵。然后,将相似度约束矩阵引入多类支持向量机求解的目标函数,构建优化的多类支持向量机分类器。最后,将优化的多类支持向量机分类器用于智能交通领域,结合梯度方向直方图特征检测行人和车辆目标。实验表明,该方法检测行人和车辆目标的错误率低于经典的多类支持向量机分类器和目前主流的目标检测方法。In order to improve the classification accuracy of multi-class support vector machine classifier for multi-target, an optimal multi-class support vector machine classifier combined with undirected graph mode is proposed. First, it uses the cosine measure to compute the similarity of training data, and builds undirected graph mode containing training data and the similarity matrix, and calculates a constrained similarity matrix. Then, it introduces the constrained similarity matrix to the objective function for computing multi-class support vector machine, and builds the optimal multi-class support vector machine classifier. Finally, it applies the optimal multi-class support vector machine classifier to the field of intelligent transportation, for detecting objects of pedestrians and vehicles with features of histogram of oriented gradient. Experiments show that the error rate for detecting pedestrians and vehicles with the new method is lower than classic multi-class support vector machine classifier and the present mainstream object detection methods.

关 键 词：无向图模型 支持向量机 多类支持向量机 相似度 目标检测 

分 类 号：TN011[电子电信—物理电子学] TP391[自动化与计算机技术—计算机应用技术]