Subdivide image correspondences into overlapping subsets with different reference views.
calibration/cg_redistribute_cors in_cors.json pseudo_refgrid.json outreach_radius out_cors.json
For example: The input in_cors.json has 2 features A and B, on 1 reference view r, both tracked over an outreach of 10:
A A A A A A A A A A A A A A A A A A A A A
B B B B B B B B B B B B B B B B B B B B B
r
-----> v
For every view v, A and B should correspond to the same scene object point. There can be an error in this because of the optical flow estimation. But for the reference view r there is no error (because the features were defined on that view).
The new pseudo_refgrid.json is given with two additional pseudo-reference views left and right of r. Then this redistribution is done with an outreach of 5. The resulting out_cors.json now have 6 features A1, A2, A3, B1, B2, B3 on 3 different reference views r1, r2, r3:
A1 A1 A1 A1 A1 A1 A1 A1 A1 A1 . . . . . . . . . . .
B1 B1 B1 B1 B1 B1 B1 B1 B1 B1 . . . . . . . . . . .
r1 r2 r3
. . . . . A2 A2 A2 A2 A2 A2 A2 A2 A2 A2 A2 . . . . .
. . . . . B2 B2 B2 B2 B2 B2 B2 B2 B2 B2 B2 . . . . .
r1 r2 r3
. . . . . . . . . . . A3 A3 A3 A3 A3 A3 A3 A3 A3 A3
. . . . . . . . . . . B3 B3 B3 B3 B3 B3 B3 B3 B3 B3
r1 r2 r3
-----> v
Here, for the reference views r1 and r3, there can now be an error in A and B (they no longer are on the right scene object point). So they are pseudo-reference views.
The new feature points A1, A2, A3, B1, B2, B3 are copies of A, B, C. So the out_cors.json gets much larger. (Binary format out_cors.bin should be used).
When used with calibration/cg_rcpos_from_cors, it will estimate relative camera positions for each of the new pseudo-reference views as center, to the error does not get as large since the outreach is smaller. Instead more error will come from stiching in calibration/cg_stitch_cameras.