Recognition of Traffic Signs based on Vision Models

This paper presents a new approach to segment and recognise traffic signs via two vision models: the colour appearance model CIECAM97s and Behaviour Model of Vision (BMV). This approach not only takes CIECAM97s into practical application for the first time since it was standardised in 1998, but also improves the accuracy of recognition of traffic signs dramatically, taking into account of changing weather conditions and varying transformations of signs. BMV model has been further developed to a foveal system for traffic signs (FOSTS). Comparison with the other colour spaces in terms of segmentation, including CIELUV and HSI, and RGB colour space is also carried out. The results show that CIECAM97s outperforms over the other three spaces with 94%, 90 % and 85% accurate rate for sunny, cloudy and rainy viewing conditions respectively. It also confirms that CIECAM97s does predict the colour appearance similar to the average observer. For recognition stage, FOSTS system gives accuracy rates up to 95%