Real-time color segmentation of road signs

Abstract

This subject is part of a more challenging research project aimed at developing a computer vision system, embedded in a car, and capable of identifying and locating road signs. Several constraints limit the possible solutions are first identified. For instance in such an application, we definitely need a real-time system; in addition, we wish to limit our choice to a single camera methodology to keep the hardware as simple as possible and finally our approach must work efficiently on low cost hardware. For this purpose we first rely on a color segmentation strategy to recognize the road signs. We have studied the behavior of the RGB components of several road signs from sunrise to sunset. From these results we have shown that the simple comparison of the RGB components taken two by two is sufficient to segment road signs in real-time. From this segmentation it should be much easier to identify the shape (and meaning) of a particular sign.