Traffic sign classification with Quantized Local Zernike Moments

Abstract

In this paper, it is shown that Local Zernike Moments (LZM), applied successfully to face recognition, can also be used successfully for the classification of traffic signs. The direct usage of the images produced by LZM is not very efficient in terms of computation time. So, a new method, named Quantized Local Zernike Moments (QLZM), is developed. QLZM has an advantage of reducing the number of images in LZM representation by packaging the binarized images of LZM. In addition, Zernike Moments (ZM) and Hue histogram are also used in conjunction with QLZM. By using only QLZM, 97.34% accuracy is achieved and by using only ZM, 93.74% accuracy is achieved. With the usage of QLZM, ZM feature vectors and Hue histogram together, 97.62% success rate is achieved. In this classification processes, the GTSRB dataset is used, which is widely preferred for the classification of traffic signs.