"The Clock" is a 24 hour hour long movie made by Christian Marclay which contains thousands of film clips each with a clock somewhere in them, so that the whole thing keeps the right time. Whilst scanning for the time in text may be tricky, the work involved in finding those clips is unimaginable. Sadly copies are only held by few museums.
I heard there were attempts to stitch together a remake with all the same scenes by some torrent site users. Not sure how far they got or if they ever finished.
The amazing progress of image recognition over the last few years would vastly reduce the amount of work required. You would need to chain together a more generic object recognition network, and feed the results into another trained network to detect numerals. Analogue clocks might work better with more traditional geometric detection.