- #1
Lindley
- 7
- 0
I am attempting to recover information about the camera motion from a series of video frames. I am referencing the book Multiple View Geometry by Hartley & Zisserman.
I have successfully computed the fundamental matrix between two frames, and extracted the second camera matrix assuming the first is the identity. However, I'm not certain whether or not this is enough to recover information about the camera motion.
I know that it won't be possible to compute the distance the camera has moved in any absolute sense, since there is a scale ambiguity. However, I would at least like to recover deltas in camera attitude and direction of motion.
The Fundamental matrix is sufficient to compute rotation and translation (up to scale) in a projective sense, but I don't understand precisely what that means or whether it's good enough for my purposes. If I consider the camera calibration to be known, I can get the Essential matrix, from which it is possible to compute 4 possible rotation/translation solutions (some of which can be ruled out by other constraints). It's probably possible to estimate the calibration matrix from the video sequence before trying to navigate off of it, but I haven't done that yet.
Is anyone here familiar with this area and able to offer insight?
I have successfully computed the fundamental matrix between two frames, and extracted the second camera matrix assuming the first is the identity. However, I'm not certain whether or not this is enough to recover information about the camera motion.
I know that it won't be possible to compute the distance the camera has moved in any absolute sense, since there is a scale ambiguity. However, I would at least like to recover deltas in camera attitude and direction of motion.
The Fundamental matrix is sufficient to compute rotation and translation (up to scale) in a projective sense, but I don't understand precisely what that means or whether it's good enough for my purposes. If I consider the camera calibration to be known, I can get the Essential matrix, from which it is possible to compute 4 possible rotation/translation solutions (some of which can be ruled out by other constraints). It's probably possible to estimate the calibration matrix from the video sequence before trying to navigate off of it, but I haven't done that yet.
Is anyone here familiar with this area and able to offer insight?