Thread: [Gpsbabel-code] height filter accuracy
Brought to you by:
robertl
From: tsteven4 <tst...@gm...> - 2013-01-14 00:09:43
|
I have change the grid used by the height filter. With the old filter a 1024 point random simulation showed these error statistics compare to EGM96-5: min error -23.683474m max error 20.074359m std deviation of error 3.360192971m mean error -0.459654324m With the new filter the same points show: min error -6.721071 max error 4.306855 std deviation of error 0.522858158 mean error 0.008354831 The new filter changes the grid interval from 10 degrees to 1 degree. The resolution is unchanged at 1m. The grid table is 63.8kB. I would prefer to use GeographicLib directly for this sort of thing. That would require license resolution and a portable way to find a geoid dataset, i.e. a file installed along with gpsbabel.exe. It would offer improved accuracy at the cost of a 2.1MB geoid dataset, but more than that we wouldn't have to test or maintain it. That would use a 15 minute grid at a resolution of 3mm, which is the least accurate grid for EGM96 that is readily available. Here is the cumulative distribution of errors from a 1,000,000 point simulation with the new grid. Large errors, e.g. ~10m, are possible but unlikely. 1 m grid 1 m grid zoom |
From: Robert L. <rob...@gp...> - 2013-01-14 06:26:32
|
On Sun, Jan 13, 2013 at 6:09 PM, tsteven4 <tst...@gm...> wrote: > I have change the grid used by the height filter. > Nice. I would prefer to use GeographicLib directly for this sort of thing. That > would require license resolution and a portable way to find a geoid > dataset, i.e. a file installed along with gpsbabel.exe. It would offer > improved accuracy at the cost of a 2.1MB geoid dataset, but more than that > we wouldn't have to > I'd prefer to not triple our binary size for people using GPSes that have goofy altitude models. Let's focus on what we're reasonably good at (coordinate conversion) and leave modeling the planet to the projects that are good at that - for the lost souls using GPSes that don't use the geoid. I'm not saying we can't grow. Let's pick our battles carefully. > Large errors, e.g. ~10m, are possible but unlikely. > Many of the GPSes in my lab will jitter 10m on altitude with very reasonable, stationary views of the sky. Now I understand why you had a height.c and a height.cc - you were caught in the machine while I was reworking everything. |