Fast traffic sign detection under challenging conditions

In recent years, a lot of researches on traffic sign detection and recognition have been done. But most of them were tested under restricted conditions such as camera with high resolution and sensitivity, highway environment or road side having a lot of trees and very few distracting objects. In this paper, we present a fast and robust traffic sign detection system including two main stages: segmentation and detection. To boost the reliability of system, a flexible segmentation stage is designed, which includes double segmentation, one with higher criteria and the other with lower criteria, to reliably cut down a great computation burden for the shape-based detection. The accuracy rate is tested to be at least 86.7% in challenging conditions, and mostly not to miss a case in usual illumination with image sequences. The dataset used in experiments is recorded with a VGA camera under diverse lighting conditions, from dark or cloudy sky to glaring condition, in urban area where a lot of confusing objects appearing on road side and target objects in few cases are partially occluded.