I have taken several series of frames that I intend to process into timelapse sequence. Some were taken using a smartphone camera in HDR mode.
Due to the post-processing that happens in the device, these images are not perfectly aligned, even if the tripod and camera are perfectly steady.
The HDR algorithm apparently does exposure bracketing, then aligns the images, crops the result to the overlapping area and scales the image back to the original resolution. This frequently causes minimal variations between consecutive images, which look slightly like minimal variations in zoom. To make matters worse, the algorithm des not even keep the horizontal and vertical scale factors in sync, thus allowing for slight distortions in images that become visible when processing them into a sequence.
With some information available on the net, I was able to piece together an instruction that works well for exposures taken under windy conditions which caused the tripod to shake, but they fail here – probably because the error here is not related to lens geometry (whose errors are well known) but a post-processing artifact, essentially a whim of some piece of software, whose correction is not purely mathematical but involves some guessing.
Any ideas on how to correct these distortions (i.e. image rectification/registration, rubbersheeting, non-rigid point set registration)? Preferrably a solution which:
- relies on FOSS tools (Hugin preferred)
- has a high degree of automation (my sequences usually have around 250 exposures, thus manually aligning images is not an option)