I’ve managed to do reverse engineering of circuit boards using GIMP and rough alignment of layers using images like the one pictured. I want to use images to reverse engineer parts in CAD, but the minor lens distortion of a camera on a phone at ~20cm high on top of a stack of objects to keep it eyeballed flat is not enough. The result is off in multiple planes. There are minor errors in my curves in the transparent CAD part pictured, but the hole pattern is correct. The picture has been calibrated to 20mm against the ruler. Any suggestions on how to make this usable for replicating the ellipse that crosses the holes ±0.05mm?
Do you have a flatbed scanner? Like a photocopier printer?
Literally put the item on the glass of the flatbed and hit scan, stick an etched mm ruler beside it facing the scanner if you want scale.
It’s generally the most effective way by far to perform this.
I’ve used it to copy the perfect profile of an extruded aluminium ebike battery housing, to print a replacement cap after the first got cracked.
Basically when you try to use your phone to perform a task like this, you’ll need to account for the inherent lens distortion of your camera, which is a pain in the ass
A decent laser etched steel carpenter’s square should run you $10-15 and give you 2 dimensions!
I see 2 problems. Both can be solved.
Position of the camera sensor plane parallel to your object’s plane.
Place a sheet of squared paper on the table for the test shot. So you can see the proper alignment in both dimensions at once. Use a tripod to fix your camera above the table.
Tell the camera to insert a little time delay (3-5 seconds) when shooting to eliminate the motion from your finger’s push.
(Edit: if you really want to stay very cheap, you could try without the tripod. But use the squared paper behind your object. Then do it all in the post processing).
Lens distortion.
Use a lens of a well known brand.(I don’t know if it is possible with a phone instead of a real camera). Then use the software “darktable” for post processing, because it has a library of lenses and can properly correct the distortion with 1 click.
If you can rig the camera so it will stay at a fixed distance from the subject, you can stitch together as many exposures as you need and it will minimize the distortion. Keep a grid in the background to help with stitching. The Google Photoscan app does something like that.
A scanner would be your best bet.
But you could go wild and ditch the photo idea.
The original piece was modeled in CAD. Simply follow the logical path for how it was originally done.
Measure the outermost dimensions and draw a rectangle as construction geometry. Then play around drawing a couple of circles - chances are that the top and bottom curves are circles and are trimmed and joined with eachother with say blend curve to preserve tangency.
Also measure the distance of the holes to the rest of the geometry. How far are they from the top and sides. You could measure their distance to eachother crosswise too to narrow down the measurement error.
The geometry is built around simple measurements, if you can work out the sketch tools and constraints, you can probably match the original with high precision.
That is what the purple overlay is already. However there are multiple ellipses that are not the same and very minor variation is important. The original part was likely made using surfacing workflow and not a parametric sketch type workflow. The error really needs to be ±0.05 or better, thus the reason it is worth chasing the picture based reverse engineering.