Since red green and blue are at different times of the scene, I wonder if you could use the differences in common visible edges between those to help smooth the "undesired" camera pose changes, and help the colors converge?
For example, say you passed a point source of light, and knew you were traveling at a constant speed. Yaw wiggle would be in difference in time between the point source being visible, and pitch wiggle would be change in sensor row where the source was seen.
And, as a consequence, I also think a depth map would pop out of the solution?
It would have been pretty easy to use a horizontal line scanner to get image speed.
In any case, it seems that using a full frame video would be much easier to post process since frame to frame correlation tells you everything you need to know about motion.
Since red green and blue are at different times of the scene, I wonder if you could use the differences in common visible edges between those to help smooth the "undesired" camera pose changes, and help the colors converge?
For example, say you passed a point source of light, and knew you were traveling at a constant speed. Yaw wiggle would be in difference in time between the point source being visible, and pitch wiggle would be change in sensor row where the source was seen.
And, as a consequence, I also think a depth map would pop out of the solution?
Basically the inverse of https://daniel.lawrence.lu/photos/ or https://trains.jo-m.ch/ ;)
The former is in the prior art section at 2:24!
I love this so much! This is such a cool hack, just for the sake of it. Amazing hacking and amazing art!
In the preso, she has said the full size output is on the website-
can anyone see what that link is? I didn't find it on the video page, but I might've just missed it.
http://philo.gay/linecam/gallery.html -- it's linked within the middle of the talk.
https://philo.gay/linecam/
It would have been pretty easy to use a horizontal line scanner to get image speed.
In any case, it seems that using a full frame video would be much easier to post process since frame to frame correlation tells you everything you need to know about motion.
One of the first scripts I vibecoded did exactly this. It works well, but is limited by framerate when there's high motion.
I'm not sure that video would have been easier to postprocess. At what fps and shutter speed?
Using the camera as a scanner has a long history in satellite cameras and other aerial photography
they need to connect with the skunkworks team, they solved this problem on sr 71 long time back !