Geometric calibration of a camera or a stereoscopic vision sensor

Glossaire

1 2 3

1.: We call ”camera” the set composed of the sensor and the associated optical system
2.: Which would be provided by an ideal camera exempt from distortion and following the pinhole model
3.: 3rd order radial distortion, 1st order decentering distortion and first order prismatic distortion

4 5 6 A-Z

4.: As seen in section "Taking the distortions into account", a physical approach to the distortion mode leads to a distortion of the points after projection on the retinal plane, before applying the transformation A which produces the discrete image (pixels map). In that case, the corrective terms are applied to the retinal coordinates, expressed in the sensor reference frame (retinal plane) rather than discrete coordinates. Compared to Figure 7, this means that the distortion function D is located between the transformation P and the transformation A (see Figure 4).
5.: The target does not need to be flat but, in practice, it is easy to draw it on a sheet of paper and to paste it on a rigid or almost plane surface. The fact that the focal point is almost flat permits easily providing an initial estimation of the parameters (X Y Z) of the calibration points M to the bundle adjustment process (see below). The method could work with any calibration object provided that an approximate 3D model of the object is known.
6.: It means that we know the values of R, R', t, t', fx, f'x , fy, f'y , cx, c'x, cy and c'y.