Flutterby™! : Calibration 2006-04-10 17:57:23.045472+00

Calibration

2006-04-10 17:57:23.045472+00 by Dan Lyke 11 comments

This is probably mostly targeted towards Eric, but: I've dug up my old Terraserver code, thinking it'd be cool to toss a GPS in my Camelbak and get some information and generate some graphics as I train, and I'm not sure my projection code is working quite right. What's a good source of landmarks easily recognizeable on topos or a few meters per pixel satellite photos with accompanying high res (a few meters) latitude and longitude information? For instance, I'd like to find a number I trust for the east peak of Mount Tam, since I can clearly pick the center of that elevation circle on the topo and see the top of the fire lookout building on the satellite imagery.

First off, buy an external antenna for the GPS if you are going to put it in your Camelbak. You can get a small one that you can sew into a hat. Further, GPS may be fun, but it's not accurate - especially in elevation. I'm sure elevation matters to you, so make sure your GPS has a barometric altimeter and calibrate it before each ride.

The USGS markers are the best possible benchmark as that's what they are designed to be. But they don't show up well in photos (being about 4 inches in diameter).

But why are you needing to reproject your data? Are you still trying to use scanned maps for your background? If so, give me an URL to the image of the entire map and I can georeference the map and reproject it into the same coordinate system as your GPS. You should be using WGS84 in your GPS.

#Comment Re: made: 2006-04-10 20:15:10.554998+00 by: mvandewettering

You can try http://seamless.usgs.gov for all sorts of information, including road data and elevation maps. I doubt you can get down to the "couple of meter" resolution that would be ideal, but they have lots of stuff available. The BTS roadmaps they have seem to be a bit more accurate than the TIGER census data I was playing with.

At this point I'm just sitting in my living room and was trying to make sure I had my numbers right for converting mouse clicks on an image map to a position, because it was looking like I was always ending up a tile west of where I thought I should. I'm using the Terraserver imagery, which is supposedly georeferenced, although every time I say that on Flutterby I get email from people who've been doing searches trying to figure out why their GPS doesn't match up with data from some source they trust.

Obviously the right thing to do is to take the GPS out for a spin and see what I come up with, but given the weather predictions (although I am getting better at reading satellite maps) that's not going to happen for a little while. Yeah, I'm painfully aware of the limitations of GPS altitude measurement (biking across the playa with a GPS will do that to ya, "wait, I should be feeling these Gs..."), and I was thinking I'd just grab elevation from DEM or NED data (if I can get the Java

interface on http://seamless.usgs.gov to work for me...). I'm not trying to keep airplanes from crashing or anything, so sampling from the 1 arc second NED data with a little bit of filtering because I'm not likely to be doing anything steeper than 20% grades should get me totally usable altitude profiles and good enough comparisons of elevation on different rides.

Us folks with other people's cheap cast-off GPS devices don't have the option of anything but WGS84... But now I'm jealous, just for the "wow, cool" factor.

WGS84 is a perfectly adequate coordinate system to use for North America (and generally adequate for the rest of the world, save Antarctica and the North Pole). And I'm surprised your GPS doesn't have a choice of coordinate systems.

How are you representing Easting? I assume decimal degrees, but are you doing "-122.3333" or "122.3333" with West implied, or are you converting to 237.66666? It may be a simple truncation problem.

And beware - most of the DEMs available from USGS are converted from the old paper quad-sheets and are about 10m. That means that you get the average elevation across a 10m x 10m square.

I'm using negative numbers for west, I'm fairly sure it's something stupid like I'm offsetting the wrong direction from the edge of the image (this is for an HTML image map), 'cause it's much further off than floating point truncation error. I'm using Proj4 and just telling it I want UTM out.

I don't think I've found any option for changing the datum, but I haven't used the GPS receiver I've got in so long that there's a good chance that I just didn't know. And since then, when I learned a little bit about datums (damn, that grates) and projections, I was fairly well convinced that all of the new data I'd be looking at would be NAD83/WGS84 (even overseas), and when I ran into something older at most it'd have been acquired in NAD27 and reprojected by the time I got it anyway.

On DEM accuracy, it looks like the stuff I can get for free, especially the stuff where the seam filtering has already been done for me, is 30m samples with +-7m accuracy. But for the most part I think I'm interested in accuracies like +-200m over 100km of distance traveled, and I'm guessing that error accumulated in one place will be dropped in the other as long as I do something smarter than point sampling. Besides, from what I've heard no two of the bike computers that use barometric pressure agree to closer than a few hundred meters after a few hours of riding, so if GPS sends me swooping up and down by fifty every now and then it's not that big a deal.

Perhaps when I have the spare time I can build the ultimate bike computer that I'm starting to think about. I could probably do reasonable things with one of those two axis solid state accelerometers to measure incline hooked into a wheel counter, to at least temper GPS, and if I find a cheap barometer chip or subassembly, just include all three.

We used a DGPS device on a project a couple of months ago for building placement and it was spec'd at 30cm (about a foot) accuracy. It's a pricey right now but watch for DGPS for the masses in the near future.

Thanks for the path stuff, I've been doing all my personal stuff on Linux recently and forgotten about Google Earth. That'd be an easy way to just slap up a "this is where I rode", as soon as it stop raining for long enough I'll have to get some tracks and start there.

The differential GPS units I've been familliar with either need additional base stations, or need to be left in one place for a while. I didn't think that accelerometer technology was to the point where it'd work on something easily portable (and I've meandered through the physics of trying to measure position with accelerometers and error accumulates really fast). Any brand names or additional info so I can look further?

DGPS implies post-processing - unless you are getting an RTK signal (real-time correction). The problem with DGPS is you only get 30cm accuracy with lots of satellites and at least a few seconds worth of samples at each waypoint. That doesn't work on a bike.

There are several good data-logging altimeters out there with very good resolution:

As well as plenty of plans. I plan to integrate one into my aerial photography rig - more for my own edification.

From what I've heard, Google's (Keyhole's) API is much better than TerraServer. Plus, you've got GoogleEarth to play with as well. It's by far one of the best geovisualization tools I've seen in a while (and that's saying something). A colleague has been having fun with a user script for ArcMap that exports to KML. The next revision of ESRI's data server, IMS, is supposed to support KML connections as well. Once that happens, GoogleEarth will have access to bewildering amounts of data (including seamless.usgs.gov).

For seams in your DEMs, let me know which ones you are using. I can mosaic them for you if you need.

For fixed extents, UTM is a better choice than WGS. They problem is going from Long/Lat in WGS to meters x/y in the right UTM zone. One problem with long/lat is that the surface distance 1 degree in longitude represents is dependent on latitude. UTM is just meters x/y from a fixed point (like State Plane).

I actually haven't gotten the chance to get try out a computer configuration where http://seamless.usgs.gov/ is more likely to work for me, but the NED files are 1 arc second processed to be seamless, so I was just going to grab those.

But for starters I refreshed my AA rechargeables, and it may actually be clear enough to ride tomorrow morning, so I'll just whip together some Perl to write a KML and drop some screen-dumps out of Google Earth.

I've got a few other microcontroller projects to do first, but once I get one or two of 'em done I'll ask Phil (now that he's a bigwig hardware developer with the ear of several chip makers) to get me some solid state barometer sensor samples. Because the idea of a GPS enabled bike computer with real battery life that can talk to the FlightDeck interface in my shifters (and use the FlightDeck buttons on the inside of my brake hoods), barometric altitude, and a solid state inclinometer sounds kinda cool!

Hmmm... How about one that has an interface to a FAT filesystem that can take a cue sheet and tell me "turn left at...". And the cue sheet rather than a map because of display power considerations.