Both Page View or Visit data could be valid depending on what you are trying to actually report on.
Using Page Views against Video Start would be basic hit level data... Number of times the page with the video was viewed and how many "starts" we done from those page views.
Using Visits on the other hand would basically de-duplicate multiple page views within the same visit... but the video starts would still potentially have multiple if people went back and re-watched / re-started the video...
You could even look at Visits against a segmented "Visits with Video Starts" so that the % is all based on visit level.
Just to complicate things, you could even look at Unique Visitors to pages with videos, and a segmented look at UVs that have "started a video" to get a percentage of users who start videos.
There are always many ways to pull stats, those ways are all valid in their own way... what really matters is what you are trying to really understand about the video usage.