mountainpenguin said:
I hadn't considered how the light source moving would affect these things. With enough training data IF the camera where attached to the light source. Given the current state of the object mapping coming from mobile systems with the IR beams combined with visible this is clearly the future it's now a question of how long before its usable and when it's going to get small enough and cheap enough for use.
It's not something I could see happening for a very long time. Basically, all object tracking at a basic level is using groups of contrasting pixels. It works great on peoples faces or cars etc because they have a lot of contrast from the area around them. However, a bit of rock in a cave is very similar to all the rock around it add in the constantly moving light source it's going to be very difficult for it to track.
Personally, I think the better path for cave surveying is some kind of lidar. Small handheld lidar is really easy to build using off the shelf components plus its a known technology in cave surveying. The issue is tracking the movement of the device between shots if that can be cracked using IMU's then the rest is easy.