ROAD-SIGNS DETECTION AND RECOGNITION IN OMNIDIRECTIONAL IMAGES

This paper describes a computer vision system for automatic traffic road-signs detection and recognition in omnidirectional images. Contrasting conventional cameras, omnidirectional ones can provide a 360 degrees field of view, which is very useful for this kind of applications. The proposed approach consists of three major steps. Firstly, the omnidirectional images are unwrapped into a panoramic form. Then traffic road-signs shapes are detected using Hu moment invariants. Finally, the recognition is performed by measuring the multi-dimensional histogram similarity between detected road signs shapes and images stored in a database representing models of road signaling panels. Experimental results have proved the performance of the proposed approach in real conditions of use.