Using Zernike Moments and SVM for Traffic Sign Recognition
Yan Wang, Chunyang Mu, Xing Ma · Journal of Highway and Transportation Research and Development (English Edition) · 2016
To prevent traffic signs from appearing in different degrees of geometric distortion in complex environments, the invariant moment, which includes translation, rotation, and scaling invariance characteristics, is used in image recognition. First, images are pre-processed. Second, the Zernike and Hu invariant moments of the images are extracted to establish the corresponding feature datasets. Third, the data set is input into a support vector machine (SVM) for target classification. Realtime-collected images and the recognition image database in the German traffic sign recognition benchmark are used in the experiment. Compared with extracting the Hu invariant moment, extracting the Zernike invariant moments and using SVM recognition both demonstrate a higher real-time recognition rate for traffic signs in a complex environment.