Occlusion-invariant tilt angle computation for automated road sign condition assessment

Computer vision techniques are being proposed and applied for road sign inspection and condition assessment to improve road safety. In this paper, we propose an automated geometric means based framework for inspection of road sign tilt angle by which it can be detected if a road sign position has been altered (i.e. tilted). The system is capable of calculating the tilt angle using geometric dimensions of road sign outline shapes. The proposed system consists of three stages: 1) color based segmentation scheme to remove background of the scene; 2) recognition of the region of interest (ROI) shape using geometric means as a road sign; and 3) tilt angle computation of road sign using geometric dimensions of the shape outline of the ROI. Implementation results using USA road signage shows that our system is invariant to scale and partial occlusion of the sign.