Projection inconsistency

[dmct]

Hello Agisfost community!

I´ve been working with A. Photoscan pro for about a year now with the topic Coastline monitoring and I am more than pleased with it.

However, on the last 2 models I´ve worked with I have noticed a projection inconsistency on my models. Both chunks have been generated with the same drone (Inspire1 and its respective camera) under the same flight plan parameters (altitude, speed, waypoints, etc., with the camera always pointing nadir), and the models too have been generated under the same alignment and dense point cloud parameters. However one model shows a somewhat more round projection than the other wich has a more planar. Inspire1 has its own GPS wich is the data I've used to georeference my models but I am not using ground control points.

So my question is, what could be generating this inconsistency and how can I fix it?

I´ve attatched screenshots of both models to make my problem more clear.

Thank you very much in advance!

Daleth C.

[frank.stremke]

looks like an allignment problem
1 its a very long narow band = bad for photogrammetry
2 it is involving water and lots of it on one side of the images its always water = bad for photogrammetry
3 camera being a very wide angle fisheye type = bad for photogrammetry and usualy results in the model shape you see if not corrected
4 on bord GPS = not realy precise enough to make god for the bad factors above
5 lack of ground control = being the main problem here since the factors above are very adverse and thats the least you can do to help photoscan do its work :-)
best regards
frank

[HMArnold]

Frank probably has a heck of a lot more experience in this field than I do, and I agree that all the things he points out except the fisheye lens on an Inspire increase the degree of difficulty.

Does an Inspire have a fisheye lens?

The primary problems that I suggest you look at are the GPS altitude and the lack of ground control points, both of which can be corrected.

For your application being on the water and you can cheat and assume that the water meets the shore at the same level all along the beach. If you generate some "synthetic" ground control points at identifiable locations along the path, the put them all in with the altitude of the lake or "0" using an import csv file, I'll bet Photoscan unbows the projection.

Assuming the mission was all flown at the same altitude relative to the home point, you can help with the terrible resolution of the GPS Altitude values by using that same import file to define the platform altitude as being the altitude of the mission for every image.

My experience is with a Phantom Pro 3 where the camera always faces forward, but it looks from the screen shot you attached that the camera might not have been pointing in the same direction as the platform. If so, Photoscan can handle that rotation, but it does add some level of complication that again can be corrected by using a csv import file to add the "Yaw" value in degrees that the camera was rotated away from the UAV heading PLUS the heading after the platform at the time of that shot.

To fly my photogrammetry missions I use Ultimate Flight by Ken Argo. It's an Android app that allows significant sophistication in mission planning, but also exports a very detailed flight log for every flight that includes GPS coordinates, the height above the takeoff point which is much more accurate than GPS altitude, as well as the craft heading (Yaw for Photoscan), the gimbal angle if other than nadir, and if I'm not mistaken for an Inspire, the angle of the gimbal off the heading.

The bottom line is that your problems are probably fixable with some synthetic GCPs and camera definition parameters using an import file

I can help if you would like

Hank
Texas
HMARNOLD@MSN.COM

[dmct]

No, Inspire1 does not have  a fisheye camera, so I suppose we can discard that posible problem. The research center I am working in recently bought a RTK GPS which we will be using for the GCP´s.

After a lot of work i was able to retrieve my  flight logs and will be using the yaw degrees to make corrections

To discard water, is masking the only and most reliable way? I am working with big chunks and a 8km coastline so that implies lots of photographic masking. Or would the GCP´s solve the problem for me?

I´ll make corrections with your recommendations and let you know how it goes. Thank you very much for your support!

[HMArnold]

I don't think you need to mask out the water.

When I have flown rectangular containment areas in the past that hold water, I have always used a normal UAV grid pattern, without GCPs or masking,  and gotten excellent results.

All images in that contain only water have no camera alignment, and as such you end up with a rectangular zone in the all-water middle where you don't get cloud points, mesh, or any orthomosaic image, but the shoreline always aligns all the way around.

Try it without masking out the water and see if your shoreline doesn't continue to align like it does now, but without the erroneous altitude curve

[dmct]

Will do and I´ll let you know how it goes. Thank you for your support!

[frank.stremke]

hi
you dont need to mask the water
you will see that the results are much better with ground control
however the shape of the object (shoreline) will still get you problems try to make the coridors not to long and increase with a bit which should get you better results. i would recomend to have at least one flightline completly over land
or have U L or T schaped flight paths with "excursions" inland of course providing ground control there so you can solidify the models poition in space a bit better and make sure they are not tilted paralel to the shoreline
frank
p.s. could you provide a sample unedited image of the inspire camera with costline ?