i scanned a little door in 15 minutes with Fujifilm X-M1 and a Zoom lens. I know, a prime lens would be better for photogrammetry. I used 23mm focal length on the zoom lens and i forgot to set the focal length in the body to the same 23mm. The bad happened as i imported the data and all images were set to 35mm. Is it correct to change it back to 23mm? I captured everything in RAW and edited the fotos in Raw Therapee and exported them as 16-Bit Tiffs without distortion correction. Because RC calculates the distortion correction automatically within the Alignment Process(with Brown 3 with tangential 2), i set the prior Image Settings (distortion, calibration and pose) to “unknown”. Is that the correct way?
Here are my Results:
Mesh:
Textured:
Camera Positions:
Bundle-Block adjustement:
I captured the Scene with 308 Photos by Hand.
Camera Settings: F7.1, Shutter speed:1/50 and ISO 200 and AF without IS ( i should’ve turned that ON).
Hi Julian,
regarding the lens distortion it’s better to let RC do the undistortion. Main reason is that a change of your focus point will also change your focal length and the distortion at the same time. This means that each image might have a slightly different lens distortion value. This effect is more visible on zoom lenses because it’s built out of more lens pieces which move when you change the focus.
I’ve never played with the priors, so I don’t know.
I’m divided about the prior undistortion with a different software. Although one always reads not to do this I am not so certain that it’s always true. Why don’t you just tra it out with your image set and post the results?
Uwe, if you don’t group the images then RC will calculate an individual distortion for each image - another strength of the software. However, it makes sense to group them first and only ungroup for a final fine-tuning.
that would be a good Idea. I think Matlab offers some calibration opportunities but it is nor for free.
Uwe you are right and thats why I won’t use a zoom lense again because the lense is moving slightly while taking images and I ended up with a slightly different focal length (between 22,4mm-23mm).
If you are near an object distortion is not so such a big deal i think, but if you are far away from it (like doing loops around an Object for better Alignment), it will make a difference. Are there Prime lenses with wide angles and little distortion? Is Brown 3 with tangential2 always the best solution in the Alignment tab?
most of my lenses are quite good about not zooming accidentaly, but even if you do it doesn’t matter much with RC, as long as you make sure to ungroup your images at the end (by standard, grouping is switched off anyway). If you want to be absolutely certain you could use some high quality duct tape to prevent anything from moving.
RC really doesn’t mind if there is a distortion in the images, because calculating them is one of it’s purposes ! ! !
If you have enough images, a perfect object, don’t zoom (or use a prime lens) and pin down the focus and everything else that can change image geometry and finally group all images, then RC will have a better result than anything from 3rd parties because each lens is unique in the tiny details. As Uwe pointed out, differing focus positions is the main reason for tiny misalignments, not zooming itself.
There is no general right answer for which distortion model is best, you need to try it out.
alright! How do I avoid the Focus issue? You mean auto focus?
I looked back to my geodetic study books about photogrammetry and there are a fue things to consider before alignment!!!
Set Prior Pose to unknown. Because you don’t know the relative Camera Position before the Alignment! You can set it to locked if you do full body scans and the Position of your cameras are therefore known.You can use alsothe options Position or Position and Orientation if the Position of your Cameras are already known and you want to finetune those Positions within the Alignment process.
Prior Calibration and Prior Lense distortion should also be set to unknown if you did not calibrated your camera by yourself and if you didn’t undistorted it in your preprocessing Raw/Image-software before importing them to RC. the option Approximate in Prior Calibration can also be used because you know the focal length you used and it will help the software to calculate it exactly. fixed will be used if you calibrated the Camera by yourself and if you don’t want RC to change anything of those values.
In the Alignment Settings I found out that the Max feature Projection Error is for detecting some Errors and to set the overall Quality/sensitivity of your Alignment.
Camera Priors i would always turn to false because Camera-GPS is weighted in the alignment process and the accuracy is pretty bad.
Am I right if I say that grouping is recommended for fixed lenses and not for zoom lenses? Because normally with fixed lenses you have nearly no movement?
Not entirely. Both fixed and zoom lenses usually have a focus. This is a group of lenses that shifts to accomplish sharp focus for different planes (in depth). This alters the image geometry slightly. When you have a zoom lens that support exif (as in that it communicates with the camera) then it doesn’t matter because RC will always group images by the focal length stored in exif, which is different at differing zoom positions.
So I would say that ideally you need a fixed lens with fixed aperture and fixed focus. But those are the old days when software could handle only one distortion for all images. It’s in the past. So depending on your setup (camera and lens, for which there are an infinite number of possibilities) it might work if you group the images even though there are slight differences. When I tested it, in my case the differences were noticeable but marginal.
you are right, I tested it and if I Group it, RC will calculate the same focal length and distortion Parameters für each image. So its more like a Mean calculation I think. thanks for advice :)! i am looking for a new DSLR with these preferences. Any recommendation ?
Fuji X-T20 (or equivalent) or Sony A7R II or III depending on the budget. Or D850, but I have no experience with that. I also like my Panasonic LX-100 - for loads of detail work it’s more than sufficient and comes with the DOF advantages of a smaller 3/4 sensor…
i would love to have a D850 but it costs 3500€ :D!!! don-t you have to take a lot of pictures because of the 12mp sensor? what lense do you use? The Sony a7rm2 costs abot 1800€ and it is nearly as good as the D850 and half of the price
If budget is an issue then I seriously recommend the Fujis. Same level in therms of noise and the only drawback is that not all the raw developers (as in DxO) can process XTrans files. But you’re obviously dealing with that already. If you need the additional pixels, then I would go for the Sony.
I use most often the 10-24 (aps-c equivalent is 15-36) mm and more often lately the 16-50 kit lens which is not bad at all. I am not too afraid of zoom lenses, as you probably notice. In my experience it is of no significance since the accuracy for my line of work is comfortably met. I maybe miss out on 1-2% but gain a lot of time since I don’t have to manually change lenses all the time. I just make sure to use the extreme settings most of the time and if something in between, then I make sure to take at least a handful of images with the same setting.
it also seems that D850 has some more chromatic aberration clearly visible on the black and white spots on the comparison frame. fujifilm is really nice and pricy and it will be a good choice for most purposes :)! I would like to the new camera mostly for capturing buildings from the ground in addition to my p4p images which have a fixed focal length of 28mm. Do you take your pictures vertically? Timothy Hanson mentioned this technique on the Webinar last Tuesday. The only benefit yout have of this technique I think is that you have more coverage from Buttom to Top.
Experience will follow - your sample is extremely good!
I usually take my images intuitively and so fast, that I sometimes have to wait for half a minute for the SD to catch up. Not a small percentage of those images are blurry because I wasn’t patient enough to keep the camera still for 1/30 of a second. It doesn’t matter, the results are still good and that is the only thing that counts (to me at least). The procedure is a means to an end…
thanks :D. It was 300 pictures and i captured pretty fast and lazy. so few pictures were pretty bad because of motion blurr caused by impaciency but RC handled it perfectly because the majority of the pictures were sharp enough. I also processed a second mesh of this object using only the closest images for mesh creation and it did not make a big difference in improvement. Did you try to scan an object with Video? I try to figure that out for p4p but i think shutter speed is too low to get acceptable results
