The Glade 4.0

This problem is CCTV/security cam related, but the solution is not necessarily specific to that field of video processing.

Backstory:

My company's ATV was stolen out of the parking garage several weeks ago. We don't know precisely when. We only have a window of 2 weeks from between the last time anyone saw it to when we discovered it was missing. The only way to get something that large out of garage would be through one of the gated exits, which have cameras. I'll be getting the building's security footage today (hopefully). It's most likely that someone loaded it onto the back up a pickup or trailer. The objective, then, is to get at least the make/model/color, and hopefully a license plate. As well, if we can find the same vehicle entering the garage, we can cross-reference the time stamp to the building's keycard records and find out whose key was used by the perp. We don't have any expectations of being able to recover it (or even get our money back in court), but we'd dearly love to nail the ******* who stole it, regardless.

TL;DR:

I have 336 hours of video to review, which even at 8x speed is 42 hours of reviewing. Watching it at something like 40x would make the review time much more reasonable, but would run too great a risk of missing what I'm looking for. Does anyone know of software that can run through a video feed and discard all the still frames? That would vastly reduce the amount of video that needs to be reviewed without sacrificing accuracy.

_________________
Sail forth! steer for the deep waters only!
Reckless, O soul, exploring, I with thee, and thou with me;
For we are bound where mariner has not yet dared to go,
And we will risk the ship, ourselves and all.