基于改进HOG-LBP特征车前多目标分类仿真

A Multi-Target Classification Method for Intelligent Vehicle Based on Improved HOG-LBP Feature

针对城区复杂交通场景,提出一种基于改进的HOG-LBP融合特征的车辆前方多目标的分类方法。在特征提取阶段,将机器学习中常用的HOG特征和LBP特征结合,并对组合特征进行改进,在不影响检测精度的同时降低特征向量的维数,提高检测速度;分类器使用二叉树支持向量机,并根据各类目标的外形特征及其在车辆前方出现的概率,设计适用于“车辆-行人-非机动车-背景”这一多目标体系的多级分类方法。实验结果表明,上述方法能较好的实现车辆前方多目标分类。

Aiming at the complex traffic scene in urban area,a multi-objective classification method based on improved HOG-LBP feature is proposed.In the feature extraction stage,the HOG feature and LBP feature commonly used in machine learning were combined,and the combined feature was improved.As a result,the dimension of the feature vector was reduced and the detection speed was improved without affecting the detection accuracy.The support vector machine was used as the classifier,and a multi-level classification system was adopted.A multi-level classification method suitable for the multi-objective classification system of“vehicle-pedestrian-non-motor vehicle-background”was designed according to the shape characteristics of vehicles,pedestrians,etc.and their probability of appearing in front of the vehicle.The experimental result shows that the proposed method can efficiently achieve multitarget classification for intelligent vehicle.