We propose a method for recognizing the ID tags of "objects" in the real world that uses the image capture capabilities of conventional cell phone cameras. The method offers robustness with regard to view-direction and view-distance. Instead of using projective transformation for geometric calibration and multi-view data, we use five colored rectangles to establish the "geometric invariants of five coplanar points". We can use the low-quality images common in the mobile environment because we use feature extraction through color calibration based on RGB values of the "Visual-Tag" background and color segmentation based on the hue information of the colored squares. Tests show that the method can yield effective ID-recognition systems.
展开▼
