Indonesian traffic sign detection based on Haar-PHOG features and SVM classification

Abstract

Abstract Segmentation and feature extraction contributes to improved accuracy in traffic sign detection. As traffic signs are often located in complex environments, it is essential to develop feature extraction based on shapes. The Haar–PHOG feature is a development of both HOG and PHOG based on Canny edge detection. One of its advantages is that PHOG feature conducts calculation in four different frequencies of LL, HL, LH, and HH. Results from experiments on four roads in Central Java and Yogyakarta using SVM classification show that the use of the Haar–PHOG feature provides a better result than the use of HOG and PHOG.