An Algorithm of Planar Coded Pattern Recognition for Camera Calibration.

We propose an algorithm for finding out the single or multiple camera calibration planar coded patterns from an image with a complicate background, and provide a kind of patterns design for multi-pattern calibration accordingly. Until recently, the camera calibration planar pattern recognition methods proposed are mainly the recognition of the calibration pattern from images with a single pattern and a simple background. Our approach expands the scope of application of object reconstruction in two aspects. The first is that the reconstruction can be operated in a natural environment instead of an arranged scene. The second is that the large objects, or objects immovable, can be reconstructed conveniently and accurately. We begin by contour detecting with geometric constraints in the given image. Then we build the contours' adjacent relations by generating planar points set Delaunay triangular subdivision. Thirdly, we obtain the correspondence between image contours and pattern elements based on the Delaunay net, pattern geometric constraints and perspective projection invariants. We demonstrate the validity of our approach by providing recognition results on real scene images.