基于Haar-like和Adaboost的车辆检测算法研究

Research on Vehicle Detection Algorithm Based on Haar-like and Adaboost

为了提高运用普通哈尔特征的Adaboost算法检测车辆的识别率,解决计算复杂等问题,提出了基于差异化较大的车辆特征区域的扩展哈尔特征,利用积分图计算特征值,通过Adaboost算法在车辆正面、背面以及侧面的样本集上分别训练出各自的分类器,并将它们组成多通道级联强分类器。通过OpenCV实现车辆检测的实验,结果证明:通过该方法能够有效地减少弱分类器数量,提高计算速度和识别率,对于实时检测视频中不同状态的车辆有较强的鲁棒性。

In order to improve the rate of vehicle detection and solve the problem of highly complexity of compute by Adaboost algorithm based on simple Haar-like,a new extension Haar-like based on vehicle’s feature region with large differences was put forward.The three different cascade classifiers were trained on front samples,back samples and profile samples by the Adaboost algorithm,then they were combined as multi-channel and strong classifier,by using integral graph to compute eigenvalues.The vehicle detect experiment was implemented on OpenCV.The results show that:it efficiently reduces the amount of weak classifier and enhances the calculating speed and detection rate.The multi-channel classifier has a stronger robustness on detecting vehicle which was appearing in the real-time video.