Vision-based Road Sign Detection

In this paper, we present a stereo-vision based approach for road sign detection. As opposed to traffic signs, which are typically made up of well-defined pictographs, road signs can contain arbitrary information. Here, color and shape are the main two cues that represent different classes of road signs, e.g. signs on the highway vs. signs on country roads. To that extent, the proposed model couples efficient low-level color-based segmentation in HSL space with higher-level constraints that integrate prior knowledge on sign geometry in 3D through stereo-vision. Additional robustness is obtained by temporal integration as well as by matching detected signs against the results of object detectors for other traffic participants. The effectiveness of our approach is demonstrated on a real-world stereo-vision dataset (3700 images) that has been captured from a moving vehicle on German highways and country roads. Our results indicate competitive performance at real-time speeds.