0

I have a camera (actually: a stereo pair) moving towards a target. While doing so the target is imaged at a constant Hz rate. From these images I would like to retrieve the motion of the cameras based on the images. The target itself is steady. At first I tried to simply compute the relative rotation and translation between the two camera poses using cameraPose in Matlab, however this seemed to give inconsistent solutions. (basically the signs would randomly flip for x&y and z). 

Therefore I decided I want to try using optical flow via opticalFlowLK and estimateFlow which results in a flow field. Obviously I am not interested in the entire flow field but rather in the motion of the camera. Is there a way to derive camera motion form the flow field? If I see this correctly I am basically looking for a motion that would result in the flow field I am computing.

Or is this a stupid attempt and my first approach was actually correct and I should continue with that? I have the feeling that the optical flow approach is very computationally inefficient since I am only interested in the motion of one camera?

Zwähnia
  • 177
  • 2
  • 8
  • Can you give some more details on the exact motion? Have you tried searching the net for *egomotion*? There are many examples, such as tracking a cars movement etc. – A. Sarid Aug 09 '16 at 17:09
  • The camera is more moving towards the target in Z direction with small movements into x and y direction (imagine a robot moving towards a textured wall). I searched for egomotion, but head the feeling that the applications are not exactly what I was looking for, often observer and the scene is moving. I am still continueing my search though! – Zwähnia Aug 10 '16 at 08:57
  • I think my main problem is, that I don't understand how I get one common velocity from the flow field I am obtaining. So fat most papers I found treat this as it would be very trivial and often simply state THAT the velocity if obtained from the flow field. My question is HOW. – Zwähnia Aug 10 '16 at 09:10
  • A bit late, but have you tried to find the 'center of expansion' and the 'time to collision'. With this two sounds you should be able to infer velocity from time to collision and small x, y corrections from the center of expansion. – martinako Aug 16 '17 at 14:52

0 Answers0