I have been reading about mono SLAM and I have a couple questions which are not so clear:
When doing pose estimation, I have seen people both reference homography transformation and also reference 5\8 point algorithm. Can homography be used? It appears it is only made to work for a planar surface but I get the impression it has been used for SLAM.
How do you maintain scale consistency between pose estimates? It's understandable that absolute scale is not recoverable but how do you make sure you maintain the same scale when estimating pose 2 from image 3 following estimating pose 1 from image 2, starting with pose 0 and image 1? Or in other words, how do you transfer scale information from the first estimate to the second?