[Geopriv] Comments on draft-thomson-geopriv-indoor-location-00

[Cullen Jennings <fluffy at cisco.com>]

I like a bunch of this draft but I think it goes just a bit too far.

I think trying to map this image just goes too far and instead we  
should map to local coordinate system in meters with arbitrary  
rotation around reference point. The mapping to the image pixel  
coordinates from this is a local matter and depends greatly on the  
type of device the image came from and it does not seem to me that we  
need to standardize that part.

Let me give some examples. When you take maps and put them on a  
scanner to digitize them, the local humidity and other issues cause  
enough variation in paper stretch that squares are not squares and  
anyone doing this seriously looks for grid lines or other reference  
points to calibrate this out. (one of the key reasons many reference  
maps are on mylar not paper) When you shoot images with a camera, you  
want a distortion model of the lenses. Lots of technology exists to do  
all this and it's commonly used but you don't want to warp the image  
as that causes image quality loss instead you map the grid onto the  
image.

Let me give a different set of examples - the calculation and  
generally approach discussed in the draft was, uh, remarkably  
Cartesian. What's going to happen at  pole? what's going to happen  
with an image that from a satellite high enough to have significant  
spherical distortion of the earth.

All of these seem like they could be easily solved by making this  
draft door the reference grid simply as a you specify a point in WGS  
84, and a rotation, and the the coordinate system for a local point  
x,y becomes move way from the reference point by a distance of length 
( [x,y] ) in a direction of the angle of [x,y] rotated by the rotation  
of the coordinate system. I probably have not said this precisely  
enough but I'm trying to define exactly what you have when the size  
are small but still make it work on a non cartesian coordinate system.

On trivial nits - I think row col are the preferred terms in an image  
when you want view row 0 as being at top. Using X,Y is sort of frowned  
on by many folks. I understand right and left hand coordinate systems  
but don't get why you think one is more convent than others. The idea  
of the integer coordinate is in the center of the pixel is certainly  
not universal - many image calibration system assume that pixel in row  
1 collects light from integer position 1 to integer position 2 putting  
the position at the edge not the center. Of course my suggestion is  
just to avoid all this stuff. I see no reason it needs to be  
standardized since.

 Cullen <in my individual contributor role>