|
I'm currently in Vagator Village, P.O. Anjuna, Bardez Taluka, Goa, India. here's a link to the area: http://www.openstreetmap.org/export?lat=15.5978&lon=73.74234&zoom=15&layers=M When I first opened the area in JOSM, I noticed a misalignment between extant OSM data and the Bing imagery by some 6-8 meters horizontally and vertically. To establish who is 'right' I gathered a few ground points from features visible on the Bing imagery, using my etrex vista hcx. I averaged the measurements for a minute each, and repeated the measurements on successive days. Conditions were ideal: clear sky, all satellites visible, no trees. My repeat measurements came out within 1m of the previous ones. Alas: if I set the image offset in JOSM to coincide with my measurements I only need a small offset for the images, and it's different to the one I would need to make the extant OSM data concide with the images. I suspect the original OSM data may have been derived from different imagery which wasn't correctly aligned. The systematic offset between the OSM data and Bing imagery can be seen well when looking at Chapora Fort: http://www.openstreetmap.org/export?lat=15.60609&lon=73.73622&zoom=17&layers=M In JOSM, an offset of 8.00; -6.00 roughly fixes the OSM/Bing difference, while for my own measurements, an offset of 1.00; 2.00 to the Bing images seems just about right. While the offset for my own measurements is so small that it's within the tolerance of the GPS measurements and therefore probably insignificant, I feel the discrepancy between the extant OSM data and both the imagery and my measurements implies that the extant OSM data are misplaced. What should I do? |
|
Given the care that you took over collecting and checking your waypoints, I suspect that they are more accurate than anything else. The fact that repeat measurements came out within 1m of the originals indicates that there was little random error in the measurements, and in the absence of trees, cliffs and tall buildings nearby, there is no reason to expect systematic errors either. It is not unusual for the georeferencing of Bing imagery to be a few metres adrift (sometimes even the alignment of different Bing zoom levels can differ by that much). It is really quite unlikely that your carefully collected and repeated waypoints are over 8m out, so we have to conclude that the existing mapping is inaccurate. This is rather confirmed by the mapping showing a track that passes through the walls of the fort in two places. Is parallax an issue? Should I chose ground points at the same altitude or does it not matter - my altitudes varying only from sea level to ca. 65m? The trail passing through the fort wall is correct - the fort is ruined and there are breaches in the wall here and there.
(22 Dec '12, 14:33)
kfj
1
You do not need to worry about parallax. A GPS receiver initially calculates its position in 3D coordinates, and converts it to latitude, longitude and height (on a specific geodetic datum, normally WGS-84) as a final step before displaying or storing the result. OSM is a two dimensional map, so it ignores the height information. Parallax correction is a problem for anyone producing a map from an aerial photograph. It can only be done if the height of the camera above the ground is known to the required accuracy. Often it is not, which causes at least some of the alignment errors.
(22 Dec '12, 22:48)
Madryn
That's a bit of a mixed message: "You do not need to worry about parallax" and "Parallax correction is a problem for anyone producing a map from an aerial photograph"!? Of course I worry only about parallactic errors in the aerial imagery. My idea was to only use ground points at the same elevation to be sure to avoid the issue altogether, but I wasn't sure if considering parallax wasn't plain paranoid in the first place ;-)
(23 Dec '12, 07:09)
kfj
2
Sorry, I was trying to convey a lot of ideas in a few words. Parallax is a problem for the people who create the Bing imagery. If it was not corrected perfectly for parallax (which is very hard to do), it will not line up precisely with accurate survey data. All that you can then do is to slide the imagery until it matches known accurate GPS traces or waypoints, before tracing anything from it. If possible, align the imagery with the average of several GPS traces or waypoints near the features that you want to map, because the Bing alignment errors may vary from place to place.
(23 Dec '12, 11:32)
Madryn
|
|
Get many GPS traces/points maybe even from different devices and at several positions (to avoid single gps reflections or shadowing) of the target area to be sure. Note that the offset could change, especially in mountainous areas. You could ask the original creators what the source of the OSM objects is (if they have no source tag and no changeset comment) and if they remember aligning the aerial imagery before tracing. And then I would move (that is what I do currently) single, small groups of streets/buidlings if I am editing there anyway and notice a strong misplacement. No moves of a whole suburb. Ensures that you do not make a too big mistake if the move was not correct for some reason. In quite some places in my mapping area I have concluded that the OSM objects are traced from misaligned aerial imagery (which e.g. the old bing (pre summer 2012) was in my area). Also see bing-imagery-problem or alignment-of-track-vs-background-imagery (found via google). 3
I found no source tag or comment. So I asked the person who entered Chapora Fort and got the answer "I did plain satellite trace and yes I did feel they are off." So what I'll do is continue re-measuring my ground points for some days (I want to do them at different times of the day) until I arrive at a satisfactorily precise offset estimate, and then I'll follow your example and move single OSM geometries which I feel are clearly misplaced.
(22 Dec '12, 14:27)
kfj
|
Unfortunately in that vicinity I don't see any useful GPS traces (in either Potlatch 2 or JOSM), so it's going to be difficult to judge.
I'll upload my tracks in a while. I mostly do footpaths, which isn't so good to crossreference with aerial imagery, since the paths here shift and change with every season.