1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
|
The test and validation tools used in the development of the b-spline
interpolated models live here.
First I wanted to come up with a surface parametrization scheme. I wanted to
project the full almost-360deg view into some sort of 2D domain that is
cartesian-ish. Then I would grid this cartesian 2D domain with a regular grid of
control points, and I'd spline that. But does such a mapping exist? What is it?
The
show-fisheye-grid.py
script exists for this purpose. It takes in an existing model, grids the imager,
unprojects the points, and projects the vectors back to a number of common
fisheye projections. I applied this to some wide models I had (their inaccurate
OPENCV8 fits), and observed that the stereographic model works well here: a
cartensian grid in the imager produces a cartesian-ish grid in the stereographic
projection. The other projections look like they'd work also. The pinhole
projection most notably does NOT work; its projections run off to infinity at
the edges.
Then I needed to find a surface representation, and the expressions. I'm using
b-splines: they have nice continuity properties, and they have nice local
support: so my jacobian will remain sparse. The
bsplines.py
script derives and validates the spline equations. fpbisp.py is a part of that.
Then I implemented the spline equations in mrcal.c, and wanted to make sure the
behavior was correct. This is done by the
verify-interpolated-surface.py
script. It produces a random splined model, projects it with mrcal, and
visualizes the sparse control point surface with the dense sampled surface. The
sampled surface should appear smooth, and should be guided by the control
points. This tool is the first test that the mrcal projection is implemented
correctly
|
