基于混合预测模型的交通标志识别方法

Traffic Signs Identification Based on Mixed Forecasting Model

随着高级驾驶辅助系统(advanced driver assistance systems,ADAS)和无人驾驶技术快速发展,交通标志识别方法成为一个重要的研究方向。为了辅助驾驶员实现安全驾驶,减少交通事故的发生,将AdaBoost-SVM和卷积神经网络(convolutional neural network,CNN)相结合,构建一个混合预测模型(mixed forecasting model,MFM),通过该模型识别交通标志。将卷积神经网络作为可训练的特征提取器,AdaBoost-SVM作为识别器。采用卷积层和下采样层构建两组隐层结构,将预处理的图像作为CNN模型的输入,利用反向传播算法(backpropagation,BP)对CNN模型进行训练直至收敛,最后将测试集的高维特征提取出来,采用AdaBoost-SVM分类器进行分类识别。实验结果表明,该MFM对交通标志具有很高的识别率和鲁棒性,且识别率和收敛时效都优于其他传统算法,对提高辅助驾驶和无人驾驶的安全性具有重要意义。

With the rapid development of Advanced Driver Assistance Systems(ADAS)and unmanned technology,the identification of traffic signs becomes an important research topic.In order to assist the driver to achieve safe driving and reduce the possibility of traffic accidents,a Mixed Forecasting Model(MFM)based on AdaBoost-SVM and CNN was proposed to identify traffic signs.In MFM,the CNN is used as a trained feature extractor and AdaBoost-SVM is used as a recognizer.The two layers of hidden layer structure were constructed by convolution and subsampling.The pretreatment images were used as the input of CNN model.The CNN model was trained by Back propagation(BP)until convergence.Finally the test set of the dimensional features were extracted,and the AdaBoost-SVM classifier was adopted to classify and identify the dimensional features.The experimental results show that the MFM has high recognition rate and robustness to traffic signs,and the recognition rate and convergence time are superior to other traditional algorithms,which are of great significance to improve the driving and unmanned safety.