Research On Algorithm For Traffic Sign Detection And Recognition

Real-time perception of road environment is one of core scientific problems in un-manned driving, and real-time detection and recognition of traffic signs in dynamic scenes is the most important in perception. This thesis researches into algorithms for real-time traffic sign detection and recognition. A system for traffic sign detection and recognition in different conditions of weather, illumination, angle, positions and background is designed and realized with high recognition rate, real-time, and robustness. The main contents are as follows:1. Based on acquisition of regions of interest of input image and types of their color and shape, comparism and analysis of the three feature extraction algorithms of2DPCA, Gabor/2DPCA, and inner pictogram extraction of the traffic sign were carried out, thus a new algorithm suited this research subject according to their merits and drawbacks is proposed. This algorithm is implemented by two steps of recognition:(1) Gabor/2DPCA and Euclidean distance,(2) inner pictogram extraction and Euclidean distance. The fusion algorithm improves the recognition rate, decreases the error identification rate, and reinforces the reliability of the system.2. A algorithm for sample library establishing is put forward. The algorithm automatically selects samples of intra-class distance, reduces quantity of samples, improves classification performance of samples and avoids randomicity of manual sample selection.3. A real-time system for traffic sign detection and recognition is designed and coded, and the comprehensive performance of the system is evaluated. Experimental results show that the average recognition rate of the system is up to96.85%, and the average process time is58.38ms. Such system is capable of detecting and recognizing traffic signs with high accuracy, real-time and strong robustness.