Recognition of multiple traffic signs using keypoints feature detectors

Automatic traffic sign recognition by computers is becoming widely desirable in reality. Methods of automatic traffic sign detection are used in the automotive industry, not only in prototypes of automotive cars, but also in mass-produced models and mobile devices. In this paper, a two-phase algorithm based on key points feature detectors to detect and recognize road signs will be presented. The first stage of the algorithm locates objects present in the scene and determines their shape based on geometric properties. In order to reduce the number of found objects first phase includes two additional steps to remove too large and too small objects, and to merge objects of the same shape found in a similar area of the scene into one object. The second phase involves proper comparison of identified object with road signs from the knowledge database based on detected keypoints. Proposed additional improvement of the algorithm is a division of characters in the knowledge base to subgroups of the same shape. Thanks to this approach, the number of comparisons between found objects of a given shape is limited to the number of signs of detected contour. Additionally, results of comparison of different algorithms that detect characteristic keypoints in a class of images that contain road signs to select proper keypoint detector will be shown.