I just wanted to know if scale is conserved across multiple images, when doing a process such as monocular odometry. I know in reality the scale tends to drift over time due to accumulation of small errors, mismatching of features and other problems. However, if I had a perfect system with perfect correspondences and perfect relative pose estimation would the scale also drift over time.
While working through the problem I believe it should not as long as each pair of images has correspondences in common with the previous image pair. However, I would like if someone could just confirm this theory of mine. Also if you have any proof as to why this is or a paper that explains it in depth that would also be very helpful.