Hi again @jlipinski,
That may be the problem indeed.
First issue states that your satellite crashed (or almost) on the Earth. Second one could be fixed with a lower minimal step but it tends to indicate there is a problem with either the measurements or the model.
Try to de-activate SRP and drag first for this OD on both days and see what you end up with.
Since your satellite is navigating between approximately 640km and 870km of altitude, the atmospheric drag should not have a big effect on a one day propagation.
This is actually quite an elliptical orbit you’re dealing with (compared to classical LEO circular orbits that have an eccentricity 10 to 15 times smaller than yours).
Beware that TLE with SGP4 is a mean model so comparing a point to an osculating orbit can be misleading.
See this post from @luc for more on that.
How did you convert camera measurements to PV in ITRF ?
Wouldn’t it be better to directly use the measurements from the camera (like RADEC or AZEL ?).
Orekit has AngularRaDec and AngularAzel measurements available.
Something you could try to evaluate the quality of OD is getting and plotting the residuals.
There are pointers on how to do it in this recent post.
This one happens to be on OD with RADEC measurements so you may find some ideas about this too.
I don’t think they are, you managed to perform an OD on both batches and make it converge to an orbit close enough to what you expected. If your observations were completely wrong, that would never have happened.
That being said, your observations may be of poor quality and that’s why you cannot follow your object properly on a daily basis.
Beware that angular measurements will badly model the shape of the orbit (a, e and ω) on such short arcs.