Research On Recognition For Traffic Sign And Platform Construction

In order to make transportation more efficient and secure, the domain of transportation tends to be intelligent. Traffic sign recognition is attached more and more attention as the core of Intelligent Transportation System (ITS). Many domestic and foreign universities and research institutions put into this investment and have had a certain amount of outputs. Traffic sign recognition plays great role on high-tech fields such as UGV etc.This thesis focuses on Traffic sign recognition, including feature extraction, data dimension reduction, SVM training and sample optimization. This thesis presents a sample optimization method based on RkNN, which enhances the real-time effect of SVM with recognizing accuracy remaining unchanged.The thesis makes initial filtration of sample database to reject some neglig-ible samples such as unrecognized traffic signs. Then selects HOG as the traff-ic sign feature after making contrast among several extractive features. Before being imported to SVM sorter, the feature is dealt with PCA and LDA dimen-sion reduction. According to the result, combining HOG feature with PCA di-mension reduction is a better way to make training and recognition.In terms of big data samples, real-time performance is not satisfying, so this thesis presents a sample optimization method based on RkNN can filtrate certain numbers of key samples from enormous database to shorten the training and recognizing time of SVM by 60%.Finally, this thesis builds traffic sign recognition application platform based on MFC and MATLAB. Making Introduction of MATLAB and C++ mixed pr-ogramming method. The platform provides a study application with integrating matured algorithms of pattern recognition and process.