Improvement of traffic sign recognition by accurate ROI refinement

This paper presents the accuracy effect of the detected traffic sign region to the traffic sign recognition (TSR) and an improved TSR method is proposed by the accurate traffic sign region extraction. The conventional HOG-based traffic sign detection (TSD) shows limited localization accuracy. Inaccurate traffic sign region affect to the TSR performance directly. For the specific TSR, speed limit signs are used to analyze the effect of localization error. Enhanced color channel based traffic sign extraction in TSR learning and testing stage validates the upgraded TSR performance.