一种新的智能车辆前方障碍物识别方法研究  

ON NEW INTELLIGENT APPROACH FOR RECOGNIZING OBSTACLE AHEAD VEHICLES

作　　者：杨镇宇[1] 黄席樾[1] 沈志熙[1] 杜长海[1] 李建科[1] 

机构地区：[1]重庆大学自动化学院,重庆400044

出　　处：《计算机应用与软件》2010年第7期17-19,29,共4页Computer Applications and Software

基　　金：国家自然科学基金项目(69674012)

摘　　要：针对现有AdaBoost-SVM(Support Vector Machine)算法中训练轮数和核函数参数选取困难的问题,以及单一核函数无法兼顾学习能力和泛化能力的缺点,提出一种基于混合核函数的支持向量机分类算法——AdaBoost-MK-SVM,并应用于城区交通干道上前方障碍物的分类识别。该算法将混合核函数作为SVM的核函数,并结合AdaBoost对核参数进行自适应调整,从而得到一组弱分类器,然后将这组弱分类器加权组合得到一个强分类器。实验结果表明,该算法能有效地对城区交通环境下车辆前方障碍物进行分类识别,分类精度高,实时性好,具有一定的优越性。Existing AdaBoost-SVM（Support Vector Machine） algorithm is difficult to select learning cycles and kernel-function parameters,and single kernel-function also has the incompatibility problem in abilities of both learning and generalization.To solve these problems,a SVM classification algorithm based on mixture of kernels named AdaBoost-MK-SVM is presented and been applied to the classification recognition of frontage obstacles on urban traffic artery.To obtain a set of weak learners,this algorithm uses mixture of kernels to be the kernel-function of SVM and combines AdaBoost to make adaptive modulation for kernel-parameters,and then this set of weak learners are weighted and combined to gain a strong learner.The result of experiment indicates that this algorithm can effectively classify and recognize obstacles in front of vehicles in urban traffic scenario and is superior to a certain extent in high accuracy and real time capability of classification.

关 键 词：障碍物识别 多类分类 支持向量机 混合核函数 ADABOOST算法 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]