Photogrammetry Scanner
Posted by leadinglights
|
Photogrammetry Scanner November 01, 2023 12:54PM |
While waiting for parts and/or inspiration that will solve the many problems on my printer rebuild I have segued back into this project which I set aside many months ago. The problem at that time was that, despite much reading and the most careful selection of components, the results I got were very disappointing.
The main problem seemed to be that the pictures lacked both the resolution and the contrast needed. The camera is a Raspberry pi HQ 12.3 megapixel one. The lens was a second-hand 16mm industrial lens of the 100lp/mm class and appeared to be pristine. Despite this, the pictures were barely as good as I get from my 15-year-old flip phone.
On picking this up again, the first try was to get a different lens, in this case, I went for a new Arducam 12mm CS-mount lens. The picture quality was very much better with this lens with the smallest lines on a focusing target being visible - with the previous lens only lines nearly 4 times as thick could be seen clearly.
This is the very first scan printed out. There are many faults and I have done almost no post-treatment of the data, but it is quite recognizable.
The rig I built is based on the remains of a Ciclop scanner and is fitted with the HQ camera and up to 6 lights (only 4 shown in photos). The lens and all lights are fitted with filters for cross-polarization. The camera is controlled by a Raspberry Pi running Linux and Picamera2
The spotlights were made from aluminium tube machined to take the polarizing filters and fitted with MR11 spotlight bulbs on the smaller lights and MR16 on the larger ones.
Much still to learn - possibly wasted effort as new AI techniques like NERF are getting so good that photogrammetry may be relegated to the buggy-whip draw; but it keeps me away from the sofa and TV.
Mike
The main problem seemed to be that the pictures lacked both the resolution and the contrast needed. The camera is a Raspberry pi HQ 12.3 megapixel one. The lens was a second-hand 16mm industrial lens of the 100lp/mm class and appeared to be pristine. Despite this, the pictures were barely as good as I get from my 15-year-old flip phone.
On picking this up again, the first try was to get a different lens, in this case, I went for a new Arducam 12mm CS-mount lens. The picture quality was very much better with this lens with the smallest lines on a focusing target being visible - with the previous lens only lines nearly 4 times as thick could be seen clearly.
This is the very first scan printed out. There are many faults and I have done almost no post-treatment of the data, but it is quite recognizable.
The rig I built is based on the remains of a Ciclop scanner and is fitted with the HQ camera and up to 6 lights (only 4 shown in photos). The lens and all lights are fitted with filters for cross-polarization. The camera is controlled by a Raspberry Pi running Linux and Picamera2
The spotlights were made from aluminium tube machined to take the polarizing filters and fitted with MR11 spotlight bulbs on the smaller lights and MR16 on the larger ones.
Much still to learn - possibly wasted effort as new AI techniques like NERF are getting so good that photogrammetry may be relegated to the buggy-whip draw; but it keeps me away from the sofa and TV.
Mike
Sorry, only registered users may post in this forum.