Detection and recognition of traffic signs in adverse conditions

Abstract

Many techniques have been developed for traffic sign recognition, but it seems related systems have hardly been applied in real vehicles. One reason is that a visible-light camera can not give competent performance in adverse conditions. In the paper, we discuss how to make the best use of a visible-light camera for over-exposure and under-exposure conditions. Two approaches are developed to enhance our traffic sign recognition system. One concerns adaptive procedures for image processing. When candidates of traffic signs are detected, their transformation to binary images and matching with templates is implemented adaptively according to their brightness distributions. Another concerns auto exposure control of an on-vehicle camera. Results of the detection component and the recognition component are accumulated temporally for several video frames, and a weighted average of them is used to pick up important regions of the current frame for traffic sign recognition. Then exposure control is performed to ensure the selected regions be reasonably bright. Initial experiment results have shown obvious improvement.