Traffic sign detection and recognition in natural scene based on human visual perception mechanism

Abstract

Traffic sign detection and recognition in natural scenes remains a challenging work for intelligence traffic system (ITS). Moreover, most traditional methods cannot achieve a better detection and recognition performance. For which some traditional methods cannot overcome the interference of complex scenes, illumination change, weather condition and rotation and scaling of traffic signs. Therefore, as the human vision perception and processing mechanisms are increasingly revealed, a traffic sign detection and recognition approach is proposed in this paper. There are mainly two parts in the proposed approach. The first part is traffic sign detection. Firstly, visual saliency detection is combined with contrast cue and center-bias cue, which is performed for detecting region of interests (ROIs) of natural scene. Moreover, the geometric structure constraint model is constructed to obtain the location of traffic signs from ROIs accurately. Next, biological-visual-based algorithm is performed in the second part, which is utilized for feature extraction of detected traffic sign. The extracted feature vector is used for traffic sign recognition through the 1-nearest neighbor. Additionally, several experiments are implemented based on Swedish traffic signs database. The recognition accuracy achieves 90%. In summary, the proposed method provides a new framework for traffic detection and recognition. For traffic sign detection it overcomes the interference of complex urbanization backgrounds. And the interference of illumination change and weather conditions for traffic signs is overcame by biological-visual-based algorithm. The contributions of this paper is significant in computational intelligence for the further work.