This paper presents a real-time speed limit sign detection and recognition system. In detection framework, only shape information was used. Firstly, the canny edge algorithm based on surround suppression was utilized, Secondly, a new contour extraction method based on topologic structure was proposed to get clearer contour, Finally, Adaboost algorithm was applied to get rid of negative candidates regions. In recognition framework, Fisher Discriminant Analysis (FDA), Bayesian rule was used to classify and validate candidate image. Extensive experimental results show that, our system has achieved detection rate and recognition rate of over 90% with a processing speed of 20-25 fps.
Samenvatting