Real-time traffic sign detection and recognition for intelligent vehicle

This paper proposes a stable system for the real time traffic sign detection and recognition, especially for the geometric distortions of traffic sign. In detection phase, color-based segmentation is applied to remove the background, then in the shape analysis subsection, the Fast Fourier transform (FFT) is used to solve the rotation and scaling problems of the traffic sign. A template database which includes the common projection distortion shapes was established to overcome the effects of the projection distortions. For object occlusions, using the method of contours convex hull to weaken the influence of occlusions. Hence, we can obtain the candidate regions of interest (ROIs). In recognition phase, the Histogram of oriented gradient (HOG) features are extracted from normalized ROIs, we propose a method which uses linear Support Vector Machine (SVM) classifier for classification. The system is verified on our intelligent vehicle named as Intelligent Pioneer. This algorithm shows good robust against scaling, occlusion, rotation and projection distortion while the accuracy of recognition is more than 93%.