Open biggzlar opened 5 years ago
This is a mono VO with the simplest method. in fact there are researches working on this problem called "scale recovery" which no need for GT, for example "Monocular Visual Odometry Scale Recovery using Geometrical Constraint". it using the ground plane detection method and estimate camera height which can be divided by the real camera height to get the scale factor.
https://github.com/uoip/monoVO-python/blob/b146da3ea25230396e51e8408f4234aba4385865/visual_odometry.py#L81
This seems weird to me. What if we don't have a ground truth (as in most applications, otherwise why do visual odometry in the first place). Any way to replace this?