Colour based road sign detection and extraction from still images

Abstract

This paper presents a system for the detection of road signs from a road scene image and extracts the pictogram inside the sign. The detection is carried out on basis of colour information of road signs instead of edge information from grayscale images. The presented colour based detection process is invariant to the common problems in road signs i.e. shadow, highlight and general affine transformation and rotation. However owing to the problems addressed in the process, it can be stated that the colour information alone is inadequate for segmentation process. Using some edge based segmentation methodology alongside would increase the detection process' performance. The idea presented in the paper is through detection of road signs with red boundaries and white interior containing black pictograms. More colours with thresholds and shapes with their Hue moments can be added without any change in the general detection methodology; colour segmentation and shape analysis. HSV colour space is adopted for colour segmentation which makes detection invariable to high-lights and illumination changes. While affine moments are used for shape analysis, which makes it invariant to rotation and scale.