1
$\begingroup$

I have a system which is composed of a rig of 8 cameras which are used for detecting markers in the environment and which outputs 8 estimates of the absolute robot's position and orientation.

Now, I need to fuse these estimations. I don't know if the best way is using a Kalman Filter or something like that.

On the other hand, I do not know if it would be convenient to track the position of each camera through a particle filter before fusing.

$\endgroup$
  • $\begingroup$ If the cameras are static, why do you want to track them? How do you calibrate the relative position of the cameras? $\endgroup$ – FooTheBar Feb 25 at 11:18
  • $\begingroup$ Yes, the cameras are static. I know their positions referred to the robot, but I want to track them referred to the world, what allows to know the position of the robot with respect to the world. $\endgroup$ – sararht Feb 25 at 11:53
0
$\begingroup$

If you just have 8 redundant transformations, I think taking an average on se3 is the simplest way. Or you can fuse them on se3 if you can estimate the uncertainty of each pose estimation. You can define the uncertainty e.g, by the distance of markers from the camera.

EKF is the best if you have complementary sensors or good estimation of uncertainty. But it seems this is not your case.

I don't see any meaning of using PF in your case.

$\endgroup$
  • $\begingroup$ Ok. In this case, does the uncertainty have to be represented in a covariance matrix? $\endgroup$ – sararht Feb 27 at 9:38
  • $\begingroup$ That's right. If you use se3 representation it should be 6x6 matrix. But I guess it will be difficult to estimate the actual 6x6 uncertainty. A simple way is to approximate the uncertainty by 1 dimension value and doing a weighted average. Please don't forget to upvote and select as answer if it was helpful:) $\endgroup$ – C.O Park Feb 27 at 22:12
  • $\begingroup$ Correction to my previous answer: We should not just take an average on se3 poses as it is on manifold. It will suffer from a singularity if the orientation of the poses is around pi. You can do something similar to the Algorithm 1 in the following paper. users.cecs.anu.edu.au/~yuchao/files/… $\endgroup$ – C.O Park Apr 6 at 16:04

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.