Visibility Estimation of Road Signs Considering Detect-ability Factors for Driver Assistance Systems

摘要

This paper proposes an automated system to estimate the visibility of road signs from the driver's perspective in daytime using in-vehicle camera images. A set of detect-ability parameters such as surrounding simplicity, occlusion, and tilting of road sign were computed to estimate the overall sign visibility. Detectability is defined as the ability of the driver to locate road sign object from a scene and thus; it measures sign posting with respect to cluttered or complex environment. The proposed system can be deployed in Driver Assistance Systems (DAS) to control information provided to drivers, since providing too much information could lead to driver distraction. Road signs are classified according to their visibility level and thus; Driver Assistance Systems can use these visibility levels to warn drivers about road signs with less visibility and high importance. The proposed system consists of four stages: 1) road sign detection and shape recognition; 2) segmentation of surrounding regions; 3) detect-ability parameters measurement; and 4) visibility level determination. This paper proposes a visibility estimation system of road signs in the United States and experimental results are used to show its effectiveness. Visibility levels from the proposed system have been compared subjectively with human expert's decisions where a notable agreement between both decisions has been gained.