摘要： 提出一种基于Boosting集成学习的二叉树支持向量机(BBT-SVM)。根据城区交通环境中各类障碍物的出现概率、模式间的类间差异，设计适用于智能车辆障碍物识别的SVM树型结构。对每个节点SVM分类器采用Boosting集成学习方法进行改进，减少差错积累误差，提高分类精度和泛化能力。实验结果表明，该方法能有效地对城区交通场景中6类常规障碍物模式进行实时在线识别。

关键词: 智能车辆, 障碍物识别, 支持向量机, 集成学习, Boosting算法

Abstract: A novel Boosting-based Binary Tree-SVM(BBT-SVM) is presented. Based on the distributing probability and pattern diversity of each obstacle in urban traffic scenes, a compatible tree structure of SVM is designed. A Boosting-based ensemble learning approach is applied to reduce the transfer error and it improves the generalization performance of per-node classifier. The improved BBT-SVM can correctly recognize six kinds of normal obstacle patterns in urban traffic scenes. Experimental results show the improved BBT-SVM can efficiently recognize six kinds of normal obstacle patterns in urban traffic scenes.

Key words: intelligent vehicle, obstacle recognition, Support Vector Machine(SVM), ensemble learning, Boosting algorithm

中图分类号: 

• TP18