In answer to my own question (below):
It seems that the "ProjectiveTransformBuilder" fails
to compute a usable transformation matrix for all
but ideal GCP configurations. I have not been able
to find the exact problem, but it seems that the
coputed transformation is only usable if source and
target point sets are each arranged in a concave
pattern and at the edges of the quadrilaterals.
I did some research and found that the projective
transform code is an old GSoC project. Has this
been abandoned? If not, then I'll happily assist
in finding the precise problem by providing some
real-worl sample data.
For now, I have fallen back to the QGIS implementation
(which works well) and wrapping that in JNI.
Best,
Ben
On 07/04/15 19:46, Benjamin Ducke wrote:
> Dear All --
>
> First of all: This is my first posting to this
> list, and I would like to take the opportunity
> and say a big "Thank You" to the developers of
> GeoTools! Your library has allowed me to do some
> immensely productive Java GIS coding.
>
> But now I've run into a dead end (or so it seems).
>
> I am trying to use GeoTools' "ProjectiveTransform(Builder)"
> classes to transform the pixel coordinates of images
> to real-world coordinates. I made some tests with
> "AffineTransform" first and this worked fine, but
> the images have a perspective distortion and a
> simple affine transformation won't suffice.
>
> Then I tested "ProjectiveTransform" with some synthetic
> data, and the results looked promising. However, with
> real data, the transformation seems to go out of control.
>
> Below is a list of six points. On line "Transform",
> you will find the image coordinates and the world
> coordinates to which they are _supposed_ to map.
>
> Under "World", you find the _actual_ mappings and the
> difference (in map units, meters). As you can see,
> the two are way off, with a maximum distance of
> 278 meters. Also note the huge RMS:
>
> PROJECTIVE FORWARD RESULTS (IMAGE TO WORLD):
> RMS = 134.46109310005252
> Pt.: 1
> Transform: 138.2/385.3 -> 513356.295/280624.993.
> World X = 513393.74146133795 (37.446461337967776).
> World Y = 280638.165346099 (13.172346098988783).
> Dist = 39.695694583763874
> Pt.: 2
> Transform: 1267.3/415.3 -> 513357.71/280624.795.
> World X = 513482.11524025456 (124.40524025453487).
> World Y = 280597.9072626728 (-26.8877373271971).
> Dist = 127.2777051229511
> Pt.: 3
> Transform: 2516.7/509.4 -> 513359.273/280624.553.
> World X = 513583.4260212551 (224.1530212551006).
> World Y = 280556.36530408286 (-68.18769591714954).
> Dist = 234.294982473546
> Pt.: 4
> Transform: 2517.4/1589.3 -> 513359.071/280623.21.
> World X = 513637.1007852404 (278.0297852404183).
> World Y = 280607.33684666886 (-15.873153331165668).
> Dist = 278.48252813687935
> Pt.: 5
> Transform: 1510.2/1744.6 -> 513357.854/280623.122.
> World X = 513573.2481027579 (215.39410275791306).
> World Y = 280647.9796511657 (24.857651165744755).
> Dist = 216.82371255092065
> Pt.: 6
> Transform: 237.0/1685.9 -> 513356.421/280623.314.
> World X = 513480.8793013462 (124.4583013462252).
> World Y = 280688.37563165755 (65.06163165753242).
> Dist = 140.43818813958055
>
> Now, I have tested the same data with QGIS and
> its projective referencing mode and it manages
> to hit the target spot-on. So the problem is not
> that the image is too distorted for a projective
> transform. The RMS should be at most a few cm.
>
> Also, if I scale the pixel coordinates to bring
> them roughly into the same range as the real-world
> coordinates (say multiply by one million), then
> I get substantially better results (however, results
> are still off the mark by several meters). This
> leads me to suspect that there is a precision problem
> involved here.
>
> I use a "ProjectiveTransformBuilder" with a list of
> "MapPosition" to build the "ProjectiveTransform". I
> have tried toying with
> "MappedPosition.setAccuracy(double)", but no luck
> (I also don't understand how to set the double
> parameter).
>
> I have tested GeoTools 11 through 13 and am getting
> the same results with each.
>
> If anyone could point me into the right direction
> for solving this problem, I would be very grateful.
> I will gladly elaborate and also send sample data
> via PM if anyone wants to take a closer look.
>
> Thanks and best regards,
>
> Ben
>
--
Dr. Benjamin Ducke
{*} Geospatial Consultant
{*} GIS Developer
Spatial technology for the masses, not the classes:
experience free and open source GIS at http://gvsigce.org
------------------------------------------------------------------------------
One dashboard for servers and applications across Physical-Virtual-Cloud
Widest out-of-the-box monitoring support with 50+ applications
Performance metrics, stats and reports that give you Actionable Insights
Deep dive visibility with transaction tracing using APM Insight.
http://ad.doubleclick.net/ddm/clk/290420510;117567292;y
_______________________________________________
GeoTools-GT2-Users mailing list
GeoTools-GT2-Users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/geotools-gt2-users
