I've been out of it for a while, but my understanding is that humans are still better at video recognition and extracting high-level information from moving videos, while the state of the art CV still tends to focus on rapid frame-by-frame static classification, with some semblance of persistence and motion strung together at the end.