Abstract:

In this paper we discuss theoretical foundations and a practical realization of a circular traffic sign detection and recognition system operating on board of a vehicle. To initially detect sign candidates in the scene, we utilize the circular Hough transform with an appropriate post-processing in the vote space. Track of an already established candidate is maintained using a function that encodes the relationship between a unique feature representation of the target object and the affine transinformation it is subject to. This function is learned on-the-fly via regression from random distortions applied to the last stable image of the sign. Finally, we adopt a novel AdaBoost algorithm to learn a sign similarity measure from example image pairs labeled either \"same\" or \"different\". This enables construction of an efficient multi-class classifier. Prototype implementation has been evaluated on a video captured in crowded street scenes. Good detection and recognition performance was achieved for a 14 class problem which reveals a high potential of our approach.