An automated road sign inventory system based on computer vision.

摘要

The overall safety provided to the driving public could benefit from an efficient inventory system that increases an agency's knowledge of existing signs. In addition, a comprehensive inventory and mechanism for updating the sign inventory could reduce the potential for liability associated with outdated, inappropriately placed, or missing signs. The key part in an automated road sign inventory system, road sign recognition, has been a challenge for computer vision. The system must recognize a wide variety of road signs under conditions of considerable variations in illumination and geometry distortion. The implementation of a system in real-time poses further difficulties.;This dissertation presents a computer vision based road sign inventory system. The goal of this development is to enhance the existing Right-Of-Way (ROW) imaging system in Digital Highway Data Vehicle (DHDV). The inventory of road sign is an application that needs to integrate multiple techniques such as image processing, stereovision and mobile mapping. The road sign inventory system is categorized into two modules: sign recognition and sign positioning. Sign recognition is conducted in three steps: detection, tracking and content recognition. The road sign region was detected primarily by color segmentation and morphological operation based shape analysis. The candidate region was tracked among successive image frames to avoid extra searching effort in each of the video frames. A predefined sign database is constructed from the ROW images. Then features invariant to scale, distortion and color were extracted for the standard signs and form a feature database. When features are extracted from the candidate sign region, they are matched to the feature database which may result in a sign recognition. In the sign positioning module, the geo-spatial information of the recognized sign on the geo-referenced ROW image was determined by stereovision model. Final results are mapped in a satellite or aerial map and stored in a relational database as well.