This is a page on my OSM activity.
How to map Lots ?
In order to avoid distortion of the stitched image far from the anchor image, I tried to add a picture of gps traces (downloaded from OSM in Merkaartor, then exported as jpg file)
Since this is a "perfect" image, I added it to the Hugin mosaic with a FOV of 1 (simulating a picture taken from very far).
Optimization now needs to be done with this picture as anchor (excluded from the roll, pitch and yaw optimisation) and as only picture optimized for the z parameter. Also disable this picture in the preview, or it might cover all others on the stitched image
I excluded the previous anchor image from any xy and z optimisation.
The average error is now 500pixels, a lot worse than before, but the resulting stitched image has only half the offset of the previous one.
Hugin is not behaving in a stable way now though. Running several optimizations will completely destroy the alignement.
I limited the number of pictures to be stitched to 14 to shorten the optimisation and stitching times (upto 3 hours on the complete set of pictures...) and make some more trials, but could not get a "perfect" result.
After a second rectification with QGIS, the offset is not much more than Google satellite, but I'm still not really happy... It's a litle useless to have a higher resolution, if the offset of the streets is not better.
I'll probably be busy the coming weeks, so I uploaded the current result here :
The same url is also valid as TMS overlay from zoom levels 11 to 21.
I suggest not to use it for real mapping, I'm still hoping to be able to improve it later. But it gives an idea of what can be achieved with non vertical pictures.
At the highest zoom level, the noise of the camera worsened by the 2 warpings becomes obvious, but you can see people walking on the streets...
Possible ways to improve the alignement:
* adding more control points, but this mean making more GPS traces.
* making a bigger GPS trace image, and testing different FOV settings
* cheating on the FOV of the other images (but when I tried values different from the actual 52, I had catastrophic results in the optimizer). The total final FOV is around 160 degrees. Rectilinear projection creates "infinite" image at 180 degrees... So I might have been on the wrong path al along when adding the GPS picture
* testing other projections
My last settings (and the GPS picture) can be downloaded here. The link of the other pictures is below.
Maayong Pasko ug Malipayong Bag-ong Tuig!
(Merry Christmas and Happy New Year)
I finally managed to update all the control points for the Hugin stitcher to align more accurately.
All the control points are placed on ground level. Mainly on the roads, using shadows of buildings and corners of buildings as reference points. I have around 30 points between adjacent pictures and some between pictures farther away to improve the general alignement.
In order to optimize correctly (this is the part where the program tries to find the parameters so that the control points are as close as possible) it has to be done in several steps:
0. add the pictures and control points. select rectilinear as projection on the stitch tab.
1 Select first only x and y alignement. Use the personalised selection, check all x and y, THEN UNCHECK x and y on your anchor image(the center image of the series, also specified as anchor in the images tab)
3. Now the images are aligned, but the angle of the camera compared to vertical is not corrected. Also if the images are not in a straight line (curved path of the camera) the stiched images will not align.
4. Now uncheck all x and y corrections, and check all Yaw, Pitch and Roll corrections. THEN UNCHECK the yaw and Roll on the anchor image (we don't want the reference to be turned around, but we need the angle correction) The yaw could be usefull if the camera was pointed slightly to the right or left, but I got weird results when leaving it checked.
5. Optimize again.
6. The imeges should now be aligned properly with only a small average error (around 1 or 2 pixels)
7. Repeating 1 and 4 in turn or together can improve further (but sometimes messes up everything...)
Now the image can be stiched (on the stich tab, auto-set the h and v field of view, auto-set the optimal image size, then set a smaller size for the first tests or stitching will take forever, then stitch )
The result : a nice looking stitched map (similar to the one below), but the 2 branches of my U are still not aligned properly (too much "open").
I think the stitching relies too much on the pitch roll and yaw to align the images, and not enough on the x and y displacement.
I'll try to add a overview picture of gps traces, to align the extremities of the U to that...
Recently I had the possibility to make a Helicopter tour over Cebu city. I took some pictures downwards to be used as aerial pictures.
Unfortunately they are not vertical, but with a pitch of 20 to 30 degrees.
This makes them difficult to rectify...
After some trials with Hugin (a panorama stitcher), I found that the last version allows to optimise not only pictures taken from a central point, but also when the camera moves parallel to the surface. Unfortunately this version is not yet in the official ubuntu repostory, but can be added as explained here.
It's a long job to add all the control points, but results are promising so far (click to open the 10mb picture):
I'm still struggling with the control points to get less distorsion. Now there are still some misalignements with osm.
More news later...
If you know about another method, please let me know ("totor_osm" on the com yahoo server).
Feel free to experiment with the raw pictures. (310Mb available here)
Map of Cebu in progress