Road-Sign Text Recognition Architecture for Intelligent Transportation Systems

Text recognition in the automotive context is a crucial task for Intelligent Transportation Systems. Its objective is to supply the driver with important information found on traffic signs. This information could be speed limits, traffic orders (Stop, for example) or texts that describe the nature of the road ahead. In this paper, a four-stage text recognition strategy is investigated. The first stage uses Histogram of Oriented gradients (HOG) features in combination with a trained suppervector machine (SVM) to detect traffic signs, specifically text-based signs such as speed-limit signs or informative-signs describing traffic situations. The detection stage is followed by a filtering stage. This stage aims to 'clean' the detected traffic sign using some filters. The filters tested in this paper are the Grayscale filter, Bilateral filter, Median filtered, and the Gaussian filler. The filtered image is then fed into the third stage, the recognition stage. An open-source Optical Character Recognition tool (OCR) "Tesseract" is used to read the texts found on the detected traffic signs. The strategy concludes with a fourth stage, i.e., post- processing, in order to add a layer of immunity to false positive and false readings. Finally, we compare our work to the standard HOG-SVM scheme. The results show that our scheme exhibits a higher accuracy over the HOG-SVM scheme.