Folks, I've had some prospective users point out known map errors, as possibly opposed to GIS-based solutions.
But it appears to me that the actual sources of the underlying map info (US, anyway) may well be limited to a small number of companies and compilations, on which the 'application-level' app's rely. Like Google uses NAVTEQ.
If this is so, then errors wd be common to all NAVTEQ-based applications. No?
Can someone provide some insight here? Like who are the suppliers of the base products and who their customers are? And who's more likely to be accurate, and why? Tnx, all.
Any errors are programmer errors. You need to verify everything before you present it to the public. Also, you need to be more specific about what type of errors, and provide a URL. Later.
> Yes, errors in the map data are errors from the data provider. People > have posted differing opinions about which provider has better data.
> AFAIK, those are the "big two" data providers in the US.
> -Ted
Being that we're using maps driven by TeleAtlas data (as opposed to Census Data and NavTeq data), then it would makes sense for those that need a solution to use TeleAtlas data for our geocoding. I'm looking into this myself now, but does anyone know if TeleAtlas offers a web service to do geocoding or if there is a geocoding web service that uses TeleAtlas data as opposed to Census data?
If they don't it would be a great way for them to generate some additional revenue over the google maps craze. I've been using the Census driven web services, and even installed it myself, but it's just not as good as I need it to be. We need TeleAtlas to publish geocoding web service, even if we have to pay for it.
Does anyone have any experience with how good of a fit their geocoding is for Google Maps vs the Census-driven geocodes?
If it fits perfectly, which one would think it should, I'm thinking that I would subscribe to TeleAtlas, for the sake of accuracy, to geocode the 2000 or so points of interest in my database. This would cost about $90 according to their site. Then run my own Census driven geocoder for when users plug in their addresses to find out where they are with respect to those points of interest to cut down on costs.
In the end this would give a solution where all of the points of interest are perfectly positioned on the google map (since both are driven by TeleAtlas), and give users the ability to roughly locate themselves with respect to those exact positions (using census driven geocode).
We began to wonder about the notion of "accuracy", and so we did a report comparing geocoder.us and the ontok geocoder (both based on Census Tiger data) to Teleatlas (Google's geocoder):
Hi Thanks for reply. I'm reading your study and noticed that you are comparing the numerical accuracy as opposed to the placement of geocoded push pins on a google map, is that correct? My concern, and it is as such simply because this is what my users will see, is that the lats and longs that are returned by whatever geocoder I use are of a value that I can use to render my push pin in the most precise location on the map.
If Main Street is numerically off by 30 feet in the teleatlas data, then it stands to reason that my push pin will end up being numerically 30 feet off if I use their data to geocode it, but with regard to what the user sees and experiences the push pin would be EXACTLY where it needs to e.
If the api maps are based on TeleAtlas data, then is it not that case that I will be able to most effectively place my push pin in the correct location using a coordinate that's generated by TeleAtlas data.
I'm just asking, but it seems to make sense. later tonight I'm going to try a couple sample points (geocoder.us vs geocode.com), though I'm not sure that I'll have the resources to generate an elaborate result set for peer review.
niyogi wrote: > We began to wonder about the notion of "accuracy", and so we did a > report comparing geocoder.us and the ontok geocoder (both based on > Census Tiger data) to Teleatlas (Google's geocoder):
Drawing Main Street (which you must do to generate the tiles) is different from pinpointing 123 Main Street (which you don't need to do to generate the tiles). Teleatlas and Navteq data are used to generate tiles of Main Street pretty effectively using the history of a GPS receiver while driving down Main Street. If they were off by 100 ft 50% of the time in where Main Street is, this would be blatantly visible in the "Hybrid" view.
What they are off on the location of *particular* street numbers on Main Street, e.g. 123 Main Street. Teleatlas and Navteq do not need to know where 123 Main street is exactly to help make the tiles. The 100ft+ errors revealed by their geocode result strongly suggests that they aren't recording many more street numbers during their data collection processes used in the Census TIGER dataset. If these street numbers were overlaid on the map, it would be blatantly visible.
I have been doing some tests with my new Holux GR-213 GPS (SiRFStarIII based): when I get really good fixes (HDOP < 1.5 with 6 or more satellites) I key them into Google Maps to compare the coordinate as shown in the resulting satellite photo with my knowledge of the terrain.
In Miami/USA, they have been spot-on (less than 2 meters of difference, if that); on the other hand, in Rio de Janeiro/Brazil, I've been getting anything from 50 to 150 meters of offset between the real and measured positions, always more or less in the SE-E direction.
At first I thought it could be due to the GPS datum, but as far as I could verify, both the GPS and Google Maps are using WGS-84. I am starting to think that Google Maps accuracy in my area (and perhaps outside of the USA in general) leaves a lot to be desired.
Has anyone done any tests like these? Care to share your results and thoughts on the matter?
What's the problem it's a map not a nuclear launch code. Get real and start using the maps for their intended use, and that is maps not a personal guide to business endeavors.
> What's the problem it's a map not a nuclear launch code. Get real and
> start using the maps for their intended use, and that is maps not a
> personal guide to business endeavors.
LOL.
Accuracy in cartography? Verifying the quality of data?
Bah. Humbug.
It's all about the JavaScript and and the pretty pictures.
Accuracy for what. Are you using for maps, or something else. Maps don't require great accuracy. just accurate enough to get a normal person where they are going. Not to put needless little pins on a map to mark commercial interests. By the way you always been a smart ass??/
That's an interesting observation. Is it possible that the local standard datum in Rio is something other than WGS-84? I'm wondering if the satellite imagery is geo-referenced to a local standard datum rather than WGS-84. If the imagery is treated as though it was geo-referenced to WGS-84 (without applying a transform), you might see systematic errors between the imagery and your GPS readings.
d_henderson wrote: > That's an interesting observation. Is it possible that the local > standard datum in Rio is something other than WGS-84?
Well, I've seen a datum called "South American 1969" referenced as a standard for Brazilian maps, and it's available to download into my GPS.
But I think that the "local standard" datum is not relevant here, _as_long_as_ both the GPS and Google Maps are using the *same* datum: I'm pretty sure the GPS is configured to use WGS-84, but I could not get an authoritative answer about Google Maps (even though I got more than one indication that they are indeed using WGS-84 too).
> I'm wondering if > the satellite imagery is geo-referenced to a local standard datum > rather than WGS-84.
You mean, Google Maps using a different datum for each country/geographic region satellite images? *Ouch* :-/ I hope that's not the case, as it would just create significant additional unnecessary complexity when using Google Maps with GPS coordinates...
I could believe Google doing that with street maps (as the originals for those could be tipically georeferenced using a local datum), but for the satellite maps I would expect the original geo-referencing to be WGS-84...
> If the imagery is treated as though it was > geo-referenced to WGS-84 (without applying a transform), you might see > systematic errors between the imagery and your GPS readings.
That is indeed what's happening, so your hypothesis above could really reflect the truth, at least in the Rio area.
I will be doing more complete verifications in the following days and test specifically for the above hypothesis, and then post the results here.
In the meantime, if anyone else has already figured this out and would be so kind as to post it here and save me a lot of sweat and aggravation, I would be grateful :-)
Other thing, i roughly remeber that GPS on south america has a accuracy of 15mts. Because earth surface variations... This is why so many datums exists! And either SIRGAS and SAD69, only append some meters to WGS84.
Situation regarding map accuracy is even worse. For large parts of the world Google Map doesn't provide any map. So we here in Europe or elsewhere have to use sat images. I remember having started a short track about their accuracy. Load http://www.druckereibeste.de/googlehupf/gpsmap.php4?sample=mw2003&dlo... and have a look at the boundary. Highways and one river are "interrupted". The greenish-looking part of the image seems to be shifted and at wrong scale.
BTW, if you look at this area through Google Earth the shift error at the river passing borders of image sources seems to reduce dramatically. Also, the brownish hight-resolution image parts cover Düsseldorf International Airport there.
Any ideas how to get correct located sat images via Google Map?
Luiz Vaz wrote: > Did you checked the values from Google Earth (GE) too?
Yes, I did. And (as I expected) I found offsets much like the ones I found on Google Maps.
> The actual Datum on Brasil must be SIRGAS not SAD69. This one is too > old, and SIRGAS replaced it on 2000.
Humrmrmrm... OK, but I don't think it's too improbable for Google to be using SAD69, as the photos of regions I know really well are at least 3 years old (based on new buildings that don't show up, etc), so maybe the datum used is even older...
Anyway, I will be checking the offsets using both SAD69 and SIRGAS. Thanks for the input.
Thanks for the pointer. BTW, do you know where I can find some tried and true formulas to convert coordinates between different datums?
> Other thing, i roughly remeber that GPS on south america has a > accuracy of 15mts. Because earth surface variations...
Really? Humrmrm... anyway, I'm detecting much bigger differences (in the range of tens of meters) so there must be some other explanations besides GPS precision.
> This is why so many datums exists! And either SIRGAS and SAD69, only > append some meters to WGS84.
By "some" you mean how much? less than 10? less than 20?
I've been checking the SIRGAS pointer you sent me, and located an interesting page with some coordinates for what I presume are geodesic reference stations:
Presuming these coordinates are georeferenced to SIRGAS datum (a reasonable presumption as they are taken from the SIRGAS site :-)), I would expect them to point straight to the geodesic station, which I would also presume to be some kind of building/construction exposed to the sky.
On a second look, maybe there's some kind of structure (white pixels) right beside those trees that I can't distinguish very well even at the max zoom level available on Google Maps. Moreover, there are some buildings there that look like some kind of campus... Hummrmrmr...
As I presume from your name that you are brazilian like me :-), I would like to ask whether you are familiar with this RIOD site. Zooming out of the picture above, I can see it's located somewhere in Rio's north zone neighborhood, apparently right besides Avenida Brasil, in the Santa Cruz -> Centro direction. Humrmrmrm... maybe I will try driving there this morning with my GPS and checking it on the spot.
Durval Menezes wrote: > I will be doing more complete verifications in the following days and > test specifically for the above hypothesis, and then post the results > here.
I've finally got some free time (yes, in the 00:00 to 01:00 AM time slot :-)) and did a more systematic attempt to find out about Google's accuracy in my area (Rio de Janeiro/Brazil).
To sum it up, there's an almost systematic difference between the "measured" and the "correct" position, on the order of 10-30 meters large, and ALWAYS more or less in the West direction; also, I've found out that the further we move East, the larger the difference gets.
I don't have any further evidence right now, but I'm starting to suspect that Google Maps isn't using the WGS84 datum, at least not in my area.
In order to investigate that, my idea is to convert the GPS coordinates from WGS84 to other datums and check if the above differences are reduced/eliminated.
Can anyone tell me a formula or algorithm to convert between WGS84 and another arbitrary datum, or refer me to some URL where this is explained? The best page I could find using Google was http://www.colorado.edu/geography/gcraft/notes/datum/datum.html, and it's far from straightforward... :-/
