Recently Lianpin Koh suggest to me to try out the Microsoft ICE. It is an advanced panoramic image stitcher. The best part : It's a freeware.
Being just completed to map 20000 Ha of oil palm land for a local plantation, I have huge data set on hand to try out if it is suitable for general use. This is a pure images stitching program, so the end result is not geo reference.
The result is simply amazing. I just load all the images and wait for the program to finish the work. No editing or intervention what so ever. Depends on the number of images, my i5 8GB ram notebook take 15 to 20 minutes to finish a 350 images data set. I'm very impress with the stitching quality, the final image looks like it was taken by a single shot.
I do encounter some problem with area cover by thick tropical rain forest. MS ICE had difficulty to stitch those images and most of the time it just ignore it and leaving the area empty. I think this is a common problem for all UAV base photogrammetry program.
The total area of this mission is about 250 Ha. The above image show smaller area because I have crop it. My Skywalker base UAV take about 30 minutes to cover the whole mission at 300 M altitude. The ground sampling distance is about 11 cm per pixel.
For those of you who just need an aerial photo of a big area, I think MC ICE can do a good job especially if the images contain many features. APM and Mission Planner is a real winning combination for aerial mapping. The system is robust and reliable. Give it a try, I'm sure you'll not disappointed.
Comments
Hi Keeyen,
thanks for the quick response, I have a canon SD1300 and A2300, I'm now learning about the CHDK to trigger the camera
Thanks
Hi Mauro
I'm using Canon S100 GPS with CHDK install. here is my UAV set up and you can have more info here. Visit the hardware and software section.
Just post here if you need further info. Wish you success in your mission.
Excellent work, I am wanting to do something, could you give me more details? which digital camera you used? how you triggered the camera, you used the command CAM_TRIGG_DIST? if you can give some details would be very useful, I intend to do my first photos next week.
thanks
For all you guys using MICE, why not give this a tryout; http://www.uni-koeln.de/~al001/airdown.html Latest version AirPhotoSE 2.44. Also has a great help file!!
"AirPhotoSE is a programme for the rectification of perspective distortion in aerial photographs. It also offers appropriate image enhancement or filtering. It can make rectified orthophotos from multiple images of the same area, construct a digital terrain model from them, and create a three dimensional interactive display of the result as well as creating a photo mosaic. It requires an image and a map or an orthophoto along with a set of matching points (control points) for each. A single non-aerial image may also be used and corrected arbitrarily to create a new image of the same size. Images from Google Earth, Google Maps, Bing Maps and national mapping service geographic portal (GeoPortal) web sites can be loaded and used instead of a map, and they can be geo-referenced automatically if the web service permits it.."
http://www.diydrones.com/m/blogpost?id=705844%3ABlogPost%3A1330059
Hi Martin, Thanks for the feed back. I have a closer look at the stitch result. It is true that there are misalignment in some area. It is not perfect yet. To me the error is small and acceptable consider it is a freeware.
I also think that the error may be due to the images input. Since the camera is attach directly to UAV, not on a stabilized mount, some images are not taken at vertical angle. This may contribute to the stitching error.
That's a hell of an area keeyen. Well done!
I tried MS ICE a long time ago, dont know if the program is better now, but the image you put in the post is not so good, lots of errors, you can see roads duplicated or trimmed.
Thanks Gerard. Good info. I'm currently using cloud service for geo ref orthomosaic. I'm more on the fly drone and take aerial image. Those post processing job will be out source. But ultimately your info will help me to understand the idea of geo ref image and their work flow. Thanks again
Keeyen,
You can georeference any image with opensource GDAL tools. They work something like this:
gdal_translate -of GTiff -gcp 0 0 -82.5586187839508 35.59414007259327 -gcp 0 410 -82.55858659744263 35.5937998255945 -gcp 520 0 -82.55741715431213 35.594340730401775 -gcp 520 410 -82.55728840827942 35.5940266570877 test.jpg test_temp.tif
TIFF is typically used, because this can contain the Georeferenced info (geotiff). It is recommended to have 5 control points in a reasonable distribution and these do not need to be corners. If you do not have GPS info for pictures, I'd use a bit more. I'd go with natural marks in your case, so you don't have to go back and recollect stuff. The preparation is going to take a bit more time though.
If you're looking at point clouds, try VisualSfM in combination with CMPMVS. These require windows with 16G memory and an Nvidia card, but the results are astounding and quite precise. You can then use the resulting .ply file in Meshlab for further analysis (volume?), especially if you used GCP, because then the units mean something.