Traffic Signs Detection and Recognition by Using a Method Based on Attention Mechanism and Classical and Support Vector Machine Classifier

Creation and developing of intelligent systems in transportation field, is one of the major issues in recent years. Detection and recognition are two main phases for traffic sign identification. Researchers have proposed different methods for implementation of these stages. This paper has presented a method for localization, detection and recognition of military traffic signs using a method based on attention mechanism and saliency toolbox. Probability of traffic signs presence in the images that are taken from traffic scenes is determined and salient regions in form of saliency maps were extracted.  As traffic signs are distinctive because of their special shape and color, they would be detected as salient regions. In the next step, for traffic signs recognition and classification, traffic signs images were divided into several groups based on their types and SIFT feature extraction algorithm were applied for keypoints extraction. A procedure was proposed to create histograms for showing signs' feature vectors for both test and train images. Traffic signs were recognized and classified by comparing of test and train histograms. Classical classifier "K nearest neighbor" with different parameters such as: (Euclidean, x2, Manhattan) and Support Vector Machine (SVM) were used for classification. In proposed method, traffic signs detection is implemented without using an extra preprocessing methods and the result was satisfactory. In spite of the variety of traffic signs images used in each category, the recognition rate is appropriate and because of using SIFT, proposed method is scale and rotation invariant and was stable to view point and illumination changes.