Please help me align depth map frames of a moving Rat

Hey guys!

I’m asking for a little help since I ran out of ideas.

I have an image stack of the underside of a rat taken with a Kinect depth sensor. Example gif here:

https://imgur.com/dJfC8SD

I already segmented the rat and aligned each frame using the centroids as anchor points. Now I would like to rotate each frame so the rat would walk in the same direction throughout the whole video.

I’ve tried co-registration algorithms to co-register every frame to the first one (or to co-register i frame with the i+1 frame while applying the transformation to 1:i frames of the stack) using the intensity based method imregtform (the feature maching ones do not work due to the lack of features) however the rotated frames become really blurry along the edges (I checked and it happens even when only translation is allowed so it’s not a result of image rotation).

I’ve tried aligning the frames based on the max. and min. feret diameters and a numerous other parameters given by regionprops (e.g. orientation) on the actual binary masks and on opened binary masks as well. I also calculated the Shortest Diameter (the length of the line which connects the two nearest boundary points and passes across the centroid) and tried aligning the frames based on that without reasonable success.

I’d like to ask for any advice or help, I feel like I’m really stuck here!

TIF stack link (200 MB): https://drive.google.com/file/d/1qHzt_5LxYtapriSFS1F-D-_TqBTAOYzO/view?usp=sharing