[Proj] Local Projection Selection

[Patrick Mézard]

I am considering points on earth surface without elevation.
I have not thought about a full fledge 3D cartesian space. However, the 
distances I have to compute are "geodesic" ones (or maybe "geoidic") not 
euclidian distances. And I assumed that approximating in a 3D space 
instead of a projective plane would be worse in term of accuracy, but I 
may be wrong. It may be faster to computer however, but I am not even 
sure since WGS84 => stereo-whatever has no reason to perform a cartesian 
space rountrip as it is done when datum shifts are required.

Patrick Mézard


Noel Zinn wrote:

> Partick,
>
> Is this a 3D problem?  Is it important for your "nearest neighbour 
> search using bounding box trees, and other things which are just 
> easier to do in an euclidian space" that the vertical dimension be 
> preserved?
>
> If so, you might consider using Earth-Centered Earth-Fixed 3D 
> coordinates or some local-horizontal or local-vertical flavor thereof.
>
> Noel
>
> ----- Original Message ----- From: "Patrick Mézard" <pmezard at gmail.com>
> To: "PROJ.4 and general Projections Discussions" 
> <proj at lists.maptools.org>
> Sent: Sunday, August 07, 2005 11:02 AM
> Subject: Re: [Proj] Local Projection Selection
>
>
>> Gerald I. Evenden wrote:
>>
>>> My immediate question is why not use a geodesic program rather than 
>>> projecting
>>> the data and using a plane system with its inherent distortions?
>>
>>
>> Because:
>> - I did not know there was code in PROJ.4 to do that and I have not 
>> much time to test/integrate new stuff. Unfortunately, it seems the 
>> geodesic computations are again using global variables everywhere to 
>> configure the ellipsoid as well as to pass and retrieve parameters.
>> - I clearly overlooked the problem, and pure distance computations 
>> are not enough. Sorry, I made you lose your time with this, I keep 
>> your remarks about the geodesic calculations. The processing code 
>> really needs the points to be projected in a plane to perform all 
>> sorts of computations like nearest neighbour search using bounding 
>> box trees, and other things which are just easier to do in an 
>> euclidian space.
>