🔒 Locked

Camera Pose estimation

Started by smythst Mar 24, 2004 at 6:32 AM 0 replies 850+ views

Original Post

Mar 24, 2004 at 6:32 AM

I'm trying to estimate the pose of a camera given an image of a calibration object. I have determined all of the points (in pixel coordinates) in the camera plane and exactly which points in the world object that they correspond to. However after this I'm quite hazy, as my lecturer's notes are pretty bad. Basically he goes about getting >> 3 points to solve for the 6 unknowns (since each point will yield two constraint equations). He then 'approximately linearises' the equations using an iterative process, and then uses the pseudo inverse of the matrix that was constructed with the linearised equations to solve the problem. Now I've been given the code to calculate the partial derivatives of the unknowns with respect to x and y, however I have no idea how to use them, or how to populate the first matrix (I'm using MATLAB by the way). Can anyone shed some light on this, or perhaps try explaining the process in *very* simple terms. Many, many thanks. [edited by - smythst on March 24, 2004 8:11:25 AM]

Conner McCloud

Mar 25, 2004 at 2:01 PM

When we did this, it was just gradiant descent. You generate a symbolic transform matrix that is made up of your six pose parameters, H, and take its Jacobian, J. Make a guess for the pose, Xi. H(Xi) is your guess as to the transform. Use your test points to get some error. J(Xi) is the gradiant of H at Xi, so update Xi by your error along that gradient.

CM

Topic Locked

This topic has been locked by a moderator. New replies are not allowed.