Stereo vision for long range up to 4 km

Question

In général stereo vision is used for robotic applications the depth is few meters. In my case i want to estimate depth up to 4 km in theory that works( large baseline 70m hight focal lenght 660 MM) fine but in expérimental i didnt try does it worth does any body try it ?? What do you think ?

Answer 1

Sounds feasible, except for a few practical caveats.

All the same math applies, just scaled up by a factor of ~1000. An equivalent would be 70 mm baseline and working range of 3.5-4.5 meters. That's comparable to the human visual system and many consumer stereo cameras.

The "usual" calibration methods (chessboard/ChArUco) won't work at that scale, so you'd have to measure and calculate the relevant matrices yourself, and physically adjust the rest.

Extrinsics:

You could point them at infinity (star-filled sky, near horizon presumably) and align such that the pictures match precisely. Then the rotation matrix would be an identity matrix. Or you could point the cameras at the same point at 4 km distance. Then you'd calculate the rotation from that distance and the baseline. For precise physical adjustment, look for "gonio stages" (goniometer).

Translation would just be the baseline distance, which is given.

Intrinsics:

Focal length can be estimated by measuring the pixel size of an object of known length. This involves a little trigonometry.

Assume no distortion, which is a fair assumption for such long focal lengths. Distortion coefficients are all 0.