gpx data vs map data

Hi guys

As you might guess i am new to OSM. I just tested my new etrex vista. I see that the gpx track is slightly beside the given road in OSM (country Senegal, Kaolack). Shall I change the road or my track accordingly ?



Yes and no. GPS-Reciever always have some offset according to weather (rain), terrain (high builings) and other conditions (solar winds aso.). So if you have nothing else than your GPS-Track to create a road: use it, it is better than nothing. I’d suggest to track a street multiple times (on different times of day and / or different days) so you can make an avarage of all the traces.

If there are any existing GPS tracks uploaded, you can realign the road, including your (then hopefully uploaded) track.
If there aren’t any existing tracks you can change the road according to your track. Anyway it would be better to go along this street another few times to make the concluding road more precise.

Usually the tracks from the GPS:es are correct, if it’s a large error ~50m or if you have lots of data the sure correct it. Otherwise it’s better to map more streets writing down names and stuff.

GPX can contain quality of the track, not sure if the etrex vista can do that (doubt it). It’s called DOP values, PDOP/HDOP etc.

I’d say that if there is only one or two tracks for an existing way and that way is within the range of GPS error, say 10m without augmentation (regardless of the position accuracy that your GPS reports - they lie… frequently) then leave it. The reason for that is that you do not have enough information to determine whether your data is any better than that of the original authors. When there is say a half dozen tracks on the same way then things become much clearer.


My experience is: One Gps track, taken with enough coverage of satellites always is exact enough to redraw an existing street, if there are differences. Many existing streets in OSM weren’t drawn on pgs tracks. Their source comes form other data like official geodata, photos etc. and are not that exact as expected.

My experience is that my high-end Garmin GPS even with more then ten satellite fixes still sometimes has large errors (> 10m). So based on one GPX track you cannot say that your track is spot on.

And in different directions; car lanes are mostly 2.5 to 4.5 meters wide; or if you’re walking with the GPS, the sidewalks on opposite sides of the road can easily be separated by a distance of 7 to 14 meters. Point being that claiming one single trace to be the one-and-only-correct position for a highway is most of the time unjustified, yet it’s what almost every road starts with.

Which can often be seen if yuo drive up and down the same road. At least with my gpsmap 60csx this gives two separate tracks!

But as said before, more tracks is better, plus knowing what it actualy looked like yourself…


Not real happy with that… so gone! Next time try I’ll to write something useful when I am awake. :slight_smile:


Lambertus wrote:

If you untrust so much your 60csx, are you then ever encouraged to enter a street to OSM based on one single track out of your gps ?

When comparing my 60csx to other tracks and many OSM streets which origine from doubtable source, I must say: gps first !


  1. In the upper left corner you see a street and my underlying trace with an angle of about 30°.
    When I saw that map the first time for me it was clear, that street is wrong, since I know it.

  2. You see the railway from left to right.
    Tracing a little street underneath the railway pretended it to be above the railway. Fortunately, there exists a public trace, obviously made by a railway passenger. Aligning the railway to that trace - and the distances were in correct order.

The map as shown here contains several severe mistakes in obviously untraced objects. Therefore: I never hesitate to realign something to my traces if I find it plausible.

I find it’s often a question of judgement. Best thing to do, as others have said, is to take more than one track and draw at the average. Accuracy is important but the vast majority of uses for this data don’t need accuracy down to the centimetre.

What Alex says. Having data is better then no data. And also there are many commercial digital maps that are off by more then 10m. Having more than one trace will be able to average the road position which is what I do all the time.

Also, I’m not saying that the 60CSx is a bad GPS, on the contrary: I can see on which side of the road I’m driving most of the time . But there are situations where it’s accuracy is not nearly what it claims to be, even with many satellite locks and good view of the sky.

I think you are arguing to much from the gps correctness of a single point. Yes, one single point may deviate for 10 m from the absolute coordinate. But is this also valid for a complete trace ?

  1. What do you guess is your overall correctness if you do the following :
    You drive a strait road (that it is strait should not be difficult to recognize) with constant speed. As result you receive an absolute strait gps trace with equidistant trackpoints.
    May it be plausible, every point to deviate 10 m or is it not much more plausible, every point to be exact within 2 m?
    And just this is my experience: my traces comply with my routes, while the drawn routes often do not: they deviate in angle (Which in gps not is possible for strait traces) and they deviate in the position and radius of the curves.

  2. Why is it possible in OSM, to transform a gps trace directly in streets?
    My personal answer: OSM has defined gps traces to be the standard in accurateness for their database. Even based on one single trace. They even do not make any directions of how to calibrate your gps. They only tell you shall trace every second.
    Therefore, I insist: It cannot be wrong to realign an existing object to a single plausible gps trace, since this is the OSM standard. Yes it is a question of judgement, but with some experience and some comparison of your traces with other traces you soon have a feeling for the accurateness of your gps. And my comparions with my gps are very good.

  3. My real experience even is as follows:
    When driving the same road in different directions I get two parallel traces which do not overlap. Therefore: it is more deciding, where the tracer has driven than the gps correctness.

That argument would be convincing, if a GPS-Receiver could have only random errors and no systematic ones. Systematic errors can move the complete track. These errors leave us with a much higher precision in the shape of the track and the distances inside of the track than with the position of the whole track.

BTW: The points in a track are not always independent measurements. I’ve found my GPS (Vista HCx) delivering even pure extrapolations as track points.


Seems now we have to go in details. What is a systematic error ?
Defect within a singular gps device.
For instance a defected transistor which only leads to a rounding error.
If your device is concerned from such behaviour its in your responsibility to verify it.
My device has proved, it has no such defect.
Proof: Look at the picture above. There you find a roundabout left at top. That roundabout and the street are quite new and have been entered to OSM by someone else. I have traced the bicycle lane around the roundabout and leading to the right for several hundred meters. It follows my trace. What do you conclude of the overall gps fault ?
I say: It is remarkable that my circle and the circle from an other gps have the same center point, which is a good indication of overall accuracy.
Algorithmic systematic gps faults.
Do we agree, that all SIRFstarIII devices are using the same algorithm for position calculation?
WouldnÂ’t it then be reasonable that a systematic error would be common to all those devices ?
I could now say why care about a systematic fault which always occurs – when drawing the map and also when driving with gps. Thus they would cancel eachb other. But there are also other systems which may have another algorithmic fault. Would you argue such faults are beyond the 1-2 m range ?

Temporary systematic faults
There are reports of position faults caused by rain, clouds, snow etc.
I cannot argue in this field, since I have no such experience. IÂ’m tracing with bicycle and that requests some good weather. I also cannot say whether they can lead to a systematic fault as you described. I think such problems are causing mainly errors in elevation measurement.
Bad satellites receive / reflections
This problem should primarily lead to strayed / missing points on your track.
Up to now I never had such problems. All my traces are smooth – even in forest - and therefore very reliable. And I consider myself able to identify a bad track if it would occur. And even if it occurs, you still can get a fine resulting road if you lay your way in the middle of the strayed points. (somewhere discribed, couldn´t find anymore)

OSM Accuracy

Weide wrote:

Please could you explain. I donÂ’t understand what you will say. I never heard a gps device will calculate trackpoints other than measure them. Why should it ? All the time when on, itÂ’s measuring and only needs to store the current value. For extra / intrapolation such algorithm needs to be implemented.
How did you verify your observation (Mapsource?)– what where your settings – Auto / Time / Distance ?

I would say that judged by my limited experience you are all right. Have a look at an example

I know where the image is from. Image is not free and it can’t be used for OSM mapping but it is guaranteed to be accurate to less than 2 m RMSE and typically RMSE is less than 1 m. I do not know who has collected GPS traces and how careful the mapper has been but I am pretty sure that the equipment has been some handheld GPS or car navigator. The result is not bad and the map is totally usable even for navigation. However, GPS measurements have lead to more than 10 m location errors in some places. Circumstances do not look very bad for GPS, there are houses with steel roof but they are low and the terrain is flat. For sure this area would have been mapped with better accuracy and faster by digitising roads on-screen.

However, Yahoo and Google have only occasionally this accurate images. They may be very sharp and pixel size may be half a meter or even less but orthorectification is usually the weak point and images have offset. I would estimate that location error in Yahoo images from Helsinki is close to ten meters on average. Error is such big that traces from handheld GPSs are more accurate and they can be used for moving objects digitised from Yahoo imagery to more correct places.

Recording traces continuously does improve the accuracy but is leaves the systematic error that is changing only slowly. I have believed that I had catched a fine, smooth and accurate trace but next day another fine smooth trace has 10 m offset compared with the first one.

Handheld GPSs and automobile navigators have been designed for navigation and they do for sure make a lot of filtering, smoothing, extrapolation and guessing. If somebody is sailing in a bad weather I guess it would really be better to show a location that is estimated by the last well known location and current speed and heading than to let the screen be blanc while GPS is acquiring new proper fix. The latter is the way how GPS units which are made for measuring work. They do not guess, because it has to be right. I have been using sub-meter class DGPS units from Trimble, Leica and Thales (each priced between 2000 and 3000 €) with external antennas mounted on a pole and they all drop the fix rather often and it can take 2-3 minutes to get it back. Garmin in a pocket is recording coordinates all the time.

My conclusion is that handheld GPSs are well suitable for OSM mapping but they do not do miracles with accuracy. And imagery with offset will lead to digitised work with offset. Accurate aerial images would be fine for digitising roads but usually they are not free. There are place for all methods.

May be this is a misunderstanding. I didn’t say, your device produces systematic errors or your data is wrong. I said, it’s wrong to estimate the precision of the measurements by just looking at how repeatable they are and how they fit together, because systematic errors cannot be found this way.

What you describe here are random errors, not systematic ones. Taking the middle of the strayed points reduces the effect of random errors; the factor of precision improvement normally is the square root of the number of points. But systematic errors lead to traces which are as smooth as the others and no one is able to recognize them by looking at the influenced data.

I agree, that bad receiving / forests primarily produce random errors, reflections however may be more complicated.

Auto and Distance in curved tunnels. The trace points fitted to a straight tunnel.

The tunnels were however relatively short, so I dont want to state, that the device records points after recognizing the complete loss of receival. But anyway, speed and direction information from previous measurements were used and so the measurements were not independent.


Yes, traveling through tunnels with a Garmin shows that individual trackpoints are averaged out using the direction vector of previous points. The results are straight tracks in tunnels with curves, corners that have a low delta direction vectors in early sections of the corner and high delta direction vectors in the mid stage of a corner to compensate (highway exits typically display this behavior). To get an idea of the average location error one can place the gpsr on a table while recording. It will immediately start to drift.

That’s odd, my GPSmap 60CSx always loses connection in a tunnel, and no track is recorded until a new fix has been re-acquired.

The drifting I have noticed too, big clouds of points on a stationary location.