TRAFFIC SIGN DETECTION USING LOCAL FEATURES

Abstract

Automatic traffic sign detection and recognition are very important for GPS-based navigation systems; however, it also raises many challenges in research and practice. Our work solves some of these difficulties: First, we have analyzed traffic sign system in real conditions in Vietnam. Besides, we have also proposed high-diversity datasets including 160 types of road signs under real-world conditions; Secondly, on using these datasets, we have done experiments based on local features and “Bag of Words” model (BoW) – which are the state-of-the-art approach in image classification and object class detection. The results are very encouraging to develop this approach in later works. Our experiments also clarify the effect of codebook size in BoW model and the drawbacks of local features.