Blur-invariant traffic sign recognition using compact local phase quantization

Traffic signs are characterized by a wide variability in their visual appearance in real-world environments. For example, changes in illumination, varying weather conditions, blurring and partial occlusions impact the perception of road signs. One of the principal causes for traffic image quality degradation is blur. This is frequently due to car motion, camera out of focus, low resolution and atmospheric turbulence. In this paper, we employ a new feature extraction method named Compact Local Phase Quantization (CLPQ) for blur insensitive traffic sign recognition. Various local descriptors such as HOG, LBP are investigated and LPQ shows a high robustness to blur. LPQ features are extracted from phase information of local regions of the traffic signs, this produces a large dimension feature vector which is not practical for real-time application. Minimum-redundancy Maximum-relevance (mRMR) feature selection method is employed to select the most discriminative and non-redundant features. Experimental results show the effectiveness of combining local phase quantization descriptor and mRMR feature selection. The proposed method achieved 98:6% average recognition accuracy on the German traffic sign recognition benchmark (GTSRB) database.