多特征融合的道路车辆检测方法

通过改进基于Haar-like特征和Adaboost的级联分类器,提出一种融合Haar-like特征和HOG特征的道路车辆检测方法。在传统级联分类器的Harr-like特征基础上引入HOG特征;为Haar-like特征和HOG特征分别设计不同形式的弱分类器,对每一个特征进行弱分类器的训练,用Gentle Adaboost算法代替Discrete Adaboost算法进行强分类器的训练;在级联分类器的最后几层上使用Adaboost算法挑选出来的特征组成特征向量训练SVM分类器。实验结果表明所提出的方法能有效检测道路车辆。

Fusing Multiple Features to Detect On-road Vehicles

Improving cascade classifier based on Haar-like feature and Adaboost, this paper proposes an on-road vehicle detection method fusing Harr-like and HOG. Firstly, HOG feature is integrated into the traditional Haar-like feature set. Additionally, different weak classifiers for HOG features and Haar-like features are designed, and Gentle Adaboost algorithm is adopted to train the layer classifiers. Finally, based on the fusion features, a cascade classifier combined with Support Vector Machine is proposed. In the last few layers of the cascade, feature vectors composed by the features that selected by Gentle Adaboost algorithm are used to train robust SVM classifiers. Experimental results indicate that the proposed method can detect on-road vehicles effectively.