Machine Learning Based Framework for Recognizing Traffic Signs on Road Surfaces

Abstract

AbstractWe propose a novel method for detection and classification of markings on the surface of the road using a camera placed inside the vehicle. Road surface markings can be defined as the text and symbol drawn on the road surface such as, stop signs, zebra crossing, pedestrian crossing, direction arrows, etc. The surface markings are differed from traffic signs which are situated on the sides of the road. These surface markings give great contribution in increasing driving safety in Advanced Driver Assistant Systems (ADAS) by providing guidance and giving the right information to driver about the markings. Their contribution also includes enhancing localization and path planning in ADAS.In our framework, we use unsupervised learning for the detection of road surface markings using clustering method and Support Vector Machine (SVM) classifier for recognizing the surface markings. Our framework performs well for almost all types of surface markings including their sizes and orientation. We have done experiments on two road surface markings datasets, dataset [17] and dataset [18] and compare it with a previous proposed method. Our experiments show that our real-time framework is robust and accurate.KeywordsRoad surface markings recognitionMachine learningComputer visionAdvanced driver-assistance system