基于后验概率SVM的交通标志识别研究

在无人车交通标志识别系统中,以传统的神经网络算法或标准的支持向量机算法(SVM)设计的分类器,只能反映样本是否属于某类而不能确定样本属于某类的可信度,提出一种后验概率SVM交通标志识别方法。首先对检测与跟踪得到的交通标志大概区域图像进行彩色分割以精确定位交通标志区域,然后采用最大类间方差法分割交通标志的内部图案,最后将分割的结果进行大小归一化作为交通标志的特征图像以训练分类器和进行识别。实验结果表明,基于后验概率SVM的交通标志识别系统在复杂的室外环境下具有很强的鲁棒性和可行性。

Research on Traffic Sign Recognition Based on Posterior Probability SVM

In the traffic sign recognition system for intelligent vehicles,the designed classifier based on traditional neural network algorithm or standard support vector machine algorithm is capable of reflecting which class the test sample is belonged to,but can not reflect the reliability.In this paper,posterior probability SVM has been utilized to design classifier.First,in order to extract the exact region of traffic sign,this paper implements color segmentation on regional image which is roughly localized by detecting and tracking step.Next,the method of Maximum Classes Square Error is implemented to segment inner diagram of traffic sign.Last,the segmented graph is normalized in a unified size as feature image and transformed to a feature vector,which is either the input for training classifier or the trained classifier to recognize.Simulation results demonstrate that the traffic sign recognition system based on posterior probability SVM is of excellent robustness and feasibility under complex natural environment.