Sign board detection is important for such computer vision applications as video surveillance and content based visual information retrieval. Previous researches on this topic focus mainly on application specific sign board such as car plates and traffic signs. Many special properties including special color, size, shape, and symmetry have to be used to detect these special sign boards. In this paper, we develop a system to detect generic sign board in an image or video. The only assumption we made is that the sign has to be a polygon formed by straight boundaries. Using a new set of straight line and polygon geometry analysis techniques, we can accurately locate a sign in a cluttered image efficiently. Experiments on over one hundred images clearly demonstrate the efficacy of the new algorithms.
展开▼
