CAMERA VIRTUAL EQUIVALENT MODEL-0.01 PIXEL DETECTORS

In computer vision, a standard approach to eliminate spatial distortion is to use a calibrated grid. We developed an original method using such a grid to establish an equivalence between any camera with inherent technical limitations (optic, solid-state sensor, digitizer) and a perfect virtual pinhole camera model. This bijection based on interpolation concepts takes all the deformations generated by the optical and electronical equipment into account (distortion, decentration, scale, and affinity ratios) without the need for their explicit determination. We propose a detection method of this grid image by the use of grey levels approximation surfaces particularly adapted to this problem. We obtained a good precision on real grid images with modelization errors standard deviation of about 0.007 pixels with an off-the-shelf 512 X 512 camera using 256 grey levels. Three interpolation methods are compared: bilinear, cross-ratio in projective coordinates and bicubic splines. We conclude from this study that, in distorted images, the use of the bicubic splines interpolation method is a good improvement.