I get some weird behavior on a turntable project: (or maybed not so weird)
First picture looks pretty good - thats how i want those tie points to be distributed
Second: is of course completely useless.
Just to make sure I understand the problem: from the pictures I would describe the problem like this:
In the First picture, the back of the book is Feature-rich enough so points are placed there
On the second picture, the features on the “walls” are more seeable for rc than the features on the book (the back of the book is no longer shown and the now visible parts of the book are not feature-rich enough)
Is that a correct assumption? Is there any “rc internal” way to get around this or do I have to change the coating of the wall ?
True, in a scenario like that you can use the DOF to your advantage and set is so that the background is slightly blurry. A larger box will also help to reduce the texture that RC can latch onto…
Thanks for the tip. I haven’t considered that Image Overlap = High is a perfect setting for turntables.
Heiko,
Just to jump on the focus wagon, your second pic looks somewhat out of focus (although it could be just a screen grab/compression artifact).
Also, it looks like you’re shooting with a fairly wide lens, which will increase your depth of field and make it harder to limit the focus area to the book. You may already know this, but just in case. If other suggestions fail, you can try shooting with a longer lens and wider aperture. That way you can limit only the close part of the book to be in focus.
Another trick could be to make the sides of your light box out of something smoother. Looks like there is quite a bit of texture on the sides that RC is grabbing on to.
Lastly, what alignment settings are you using for Max Features per Image and Preselector Features? If all else fails, you can bump those and try to force RC to find more tie points, which should hopefully result in some ties on the book.
nice tips! Didn’t think of the angle and respective DOF.
With my recent trouble project, an exterior of a partly ingrown building, I did not notice any distinct difference between high, medium or low image overlap, neither in the error report nor in the amount of images in a component. Weird.
Anyway I would suggest to try Preselector first, because that is the most important in my view because it defines how many features will actually be used. Although it might be a good idea to raise the Max Features as well because that would force RC to include less ideal features (e.g. on the book) since it picks the best ones first. Slowly, I am starting to grasp how all of this goes together…
because with this tool you dont give the choice to RC, For sure it can be tricky and time consuming with number of pictures increasing, but, very rewarding, for me it worth the pain
i did it on a project last year, i had to do a canal boat. without it RC made total S**T, because obviously the boat was moving on the water (captain obvious there )
The boat isnt there, (i erased original images it was taking too much space on my hardrive) but you have an idea of the site context. raw result 280M reduced to 10M for this pointcloud, made from 2345 pictures.
NB to RC team: i discussed this project with a huge laser survey company based in Brussels, they laughed at me at start saying i was going nowhere, but when they saw the results… overwhelmed and staring faces i guess its another win for photogrammetry against lasergrammetry
How did you do that, run along the river while the boat was moving??
Both techniques have there pros and cons, I wouldn’t say one is better than the other.
But I agree with you that photogrammetry is often still smiled (if not laughed) at. Just met a surveyor this morning and tried to tell him a bit about it. Ther was no chance though, since he “has tried it too” and it wasn’t accurate and anyway he knows better, obviously. :-) I can now confidently say that I can scan a building with an accuracy of a few mm - totally comparable to a scanner, since they aren’t flawless as well and while one individual scan is extremely accurate, there is some loss after the alignment. Not to mention all the stray points that can be quite ugly and unhelpful.
I would maybe try to fly a drone back and forth on a line perpendicular to the barge movement. This should effectively give a raster “ortho” flight pattern. Or do the same thing on the side, where the drone is just going up and down continuously in one spot while the barge moves. Or better yet, combine those and fly a semicircle from one side to another !
A lot of background fixing in post processing though. Glad that focus area trick worked out.
Yes shots over and around the boat, yes a drone flight would have been great but impossible with the actual belgian autority ("its too dangerous, they ask for so much security “shields”, making the cost impossible for a small project)
I think what’s great for photogrammetry here is that its really easy to remove area from the pictures, its seems that its really hard to remove point from a laser scan.
It’s like with everything - if you know what you’re doing you can also filter from a laserscan. What you absolutely cannot do though is to scan a moving object!
The focus area tool seems quite handy, but you would need a recognizable difference in focus, which might not be the case with a huge DOF and small distances. If Heike used F11 or beyond, I doubt there would be much to go on…
for drones and a moving object for scanning - the only option that would make a sense is to lock your drone on 1 spot and point the camera towards the moving object while taking images with sufficient overlap that might be hard to control in such situation. Take these images from several different positions in space and you might get the result. Obviously you have to mask and filter all unnecessary noise apart from the object, else you might end up in a messy model.
ill give that a try
