sour mica
Hey Now! I've dug back into my linear capture setup, with Python on Raspberry Pi handling motor controllers, camera, and rough quilting. I'm working on a total refactor of the naive code to slightly less naive, but my background is in digital imaging & not at all in code. Here's a current rough quilt for fun - I'll fine-tune the convergence next week at the library.
https://blocks.glass/lichenshare/02d9d8e5-655f-42d4-9dec-4fbc1942ea9e
flint grove
Hey Now! I've dug back into my linear capture setup, with Python on Raspberry P...
that's nuts!!!
sorry, I wanted to come to AWE but my family was visiting and the timing was too close
sour mica
that's nuts!!!
Thanks! This capture is from an ArduCam 64MP sensor, in binned + summed mode uncropped, with stock focus-controlled lens behind BeastGrip Pro Macro lens. The motor controllers are discrete microcontroller + stepper motor driver boards, currently connected over WiFi as their UART details aren't documented. Python controls the camera through picamera2 and pastes the captured views into a quilt. Having this tool has made it much easier to experiment & dial into sweet spots.
sour mica
Here's a rough converge of a more even focus plane.
The previous quilt shows significant distortion at the extents, due to the macro lens. The texture & depth range of the cloth in this image help to hide some of the distortion as noise. Vertical alignment (not yet implemented in rough convergence) will also help.
https://blocks.glass/lichenshare/e6d6a547-27f0-4dae-8190-9f38cb23c885
rose schooner
are you using like a photo rail or something?
sour mica
are you using like a photo rail or something?
Yes, I'm using a motorized rail, based on designs & parts from OpenBuilds, controlled by WiFiStepper 8266+PowerStep01 from CrowdSupply. WiFiSteppers are controlled by REST through Python on Pi.
For macro photography in particular, precise travel between shots makes a big difference.
rose schooner
its surprising to me that the linear rail causes the weird distortion. When I programmed our photo rail software I didn't run into this but I was also working at a much larger scale, I think the total rail distance I was using was like 90 cm
sour mica
The distortion is mainly due to the wide angle of the stock motor-focus lens on the ArduCam sensor, compounded by the macro lens. A much more common lens stack would include a base lens with a much longer focus.
Without the macro lens, I don't particularly see distortion with travel ~365mm (the extent of my 500mm rail/screw)
Currently I'm also using the sensor at full uncropped resolution (binned & summed), so I'm using significantly more of the frame (top & bottom) than most macro workflows.
rose schooner
ah ok, that makes sense, I wonder if you could unwarp it using opencv or something
sour mica
I'm looking at moving from opencv to scikit-image for some of those functions.
sour mica
I'm getting much less horizontal distortion from a shorter travel distance using less of the sensor width. Here are a few current takes of stones lent by a friend, still with rough horizontal & no vertical convergence. Once the weather permits, I'll put in some time at my library's makerspace on an x64/altDP box for Triaxes 3DMasterKit for fine convergence & LKG Bridge for preview.
sour mica
@rose schooner re: linear capture from video - turning bugs into features?
https://jerredchen.github.io/image-as-imu/
rose schooner
ah, thats cool! I wonder if you could use this to improve the tracking used to estimate camera poses for splats and stuff