两种智能交通标志分类器的比较研究

本文介绍了一种可用于交通标志识别的新方法--支持向量机（SVM）算法,并将SVM算法与BP算法在交通标志的粗、细分类中的识别效果进行了对比分析。用中国的116个和日本的23个交通标志标准图分别训练基于SVM算法和基于BP算法的智能分类器,并用中国标志的噪声图、扭曲图和531个日本交通标志实景图作为测试集。在粗分类中,虽然BP算法 法的识别率也能达到90%以上,但SVM算法的识别率几乎可达100%,二者差距明显。在细分类中,SVM算法的识别效果与BP算法相比具有更加明显的优势。实验研究结果表明,SVM算法可以以接近最优的方式解决模式分类问题,同时具有更好的泛化能力,在交通标志识别领域具有良好的研究价值和应用前景。

Comparative Research of Two Intelligent Traffic Sign Classifiers

Support vector machine(SVM)is a novel machine learning method based on the statistical learning theory,which can avoid over-learning and provides good generalization performance.In this research,Multi-category SVM(M-SVM)is applied to traffic sign recognition and is compared with the BP algorithm,which has been commonly used in neural networks.116 Chinese ideal signs and 23 Japanese signs are first chosen for training M-SVMs and BP intelligent classifiers.Next,noise signs,level twisted signs from real Chinese and Japanese traffic signs are selected as a test set for the purpose of two network testing.Experimental results indicate that SVM has achieved a nearly 100% recognition rate and has certain advantages over the BP algorithm in approximated classification for traffic signs.In fine classification,SVM shows its superiority to the BP algorithm.Based on the analysis of the results,one may come to a conclusion that SVM algorithm is well worth the research efforts and is very promising in the area of traffic sign recognition.