A collection of benchmark images for traffic sign research

Abstract

This paper details a collection of traffic signs that has been made available online to meet the current need of lack of standard datasets for evaluation and comparison of existing and future developed recognition systems of traffic signs, leading to the exploration of the breath of innovations and applications in the field. Examples on how to apply these data are also described with comparisons between a number of approaches on segmentations. This collection aims at improving the accuracy of sign recognition systems and expediting the processing speed. Within this remit, the images in the collection opt for 2D still pictures and are taken under controlled viewing conditions on London roads at the UK. The standard sign database employs British road signs (143 images), which are available at http://www.mitime.org/traffic_signs/, together with the collection (128 images) in this study.