They figured out how to "summarize" video surveillance footage by attaching a hovering timecode to any object moving through the frame and then later it overlays several hours of footage into one video.
If you sign up for NFL Replay, you can watch "the full 22" videos during the week. It's inexpensive. I hate that they don't show this view on regular broadcasts. It's such a big part of the game, to know what type of coverage the DBs are in when a play starts. To know whether the QB made a good read, you'd have to see the coverage on all of the receivers.
I agree. I enjoy football strategy and like seeing different coaching styles in action, but this is hard to pay attention to during a standard broadcast. So I really love the trend toward offering multiple ways to view games. If you haven't seen it already, the "BCS Film Room" edition of last year's National Championship game was one of the best broadcasts I've ever seen.
I would casually watch games in this format last season just because it's super fascinating -
However, I absolutely fell in love with it rewatching the Super Bowl (Hawks fan) and just following Kam Chancellor - The guy got completely snubbed for MVP
Here's another person's 22 man camera view talking about The Touchdown Canceller's (quite literally perfect) performance:
Edit: I don't know if they offer the All-22 but NBC's Sunday Night Football is always streamed online for free and they offer multiple camera angles. It's pretty nice compared to what other broadcasts offer.
I assumed it had something to do with not providing that information to the currently playing teams while they are currently playing, are teams allowed to have "spotters" up in the stands to get an overhead view?
"Lonnie Marts, a former linebacker for the Jacksonville Jaguars, says there are thousands of former NFL players who could easily pick apart play-calling and player performance if they had access to this film."
And another one for the conspiracy buffs:
"The NFL makes a handful of plays from the All 22 available on its web site for a fee, but they're often so blurry the players' numbers aren't visible."
Very interesting! With some manual or automatic annotation, this is one step closer to spatiotemporal player data, which opens up whole new boundaries of analysis.
As a point of comparison, there's a private NBA dataset (based on imaging and manual annotation) which shows the locations of shots made, annotated with the player name and whether they failed or succeeded - from this, you can do really cool things like Miller et. al. 2014 [1], who use dimensionality reduction to find patterns in players' preferred shot locations, potentially allowing you to create teams whose offensive abilities are complementary from a spatial perspective.
As a programmer, football player, and football coach, I've been searching for something like this for a long time.
I think the biggest advancement in technology would be figuring out what plays are being run and in what scenarios, and analyzing them for patterns and weaknesses. Like knowing that the opposite team runs GAT to the wide side of the field 95% of the time. Or figuring out individual players' tendencies to give more detailed read keys to the defense.
The problem is, I think, that a lot of the critical info is inside the tackle box with the actively engaging linemen. It seems it would be tricker to track ten or so players in such close quarters.
Very neat. It's interesting to see the NFL logo from the corner of the original footage moving around in the stabilized footage. Future feature: automated watermark removal?
As the watermark is static in the original footage, it should be easy to remove before processing.
I'd like to see the players remain, perhaps somewhat transparent, once they are cropped out of the actual shot. Otherwise the zoom/pan just erases them.
This seems like just background subtraction. I used this on soccer videos eight years ago using some robust optical flow code from Michael Black to estimate the background motion and suppress it, leaving the foreground (i.e. the player).
Fantastic Blog, I run an online football game and try to use advanced analytics to generate proper games. Going to rummage through your archive and see if there's anything usable for me, trying to generate accurate games.
Many Qbs improve or decline their QB rating, while actually being better QBs or worse depending on the yardage thrown. I could give many examples of QBs with decent Completion %s that are horrible because they throw short passes.
Not bad, just another note that distance is the foundation of completion % and without that, it's essentially bad data.
One interesting analysis would be plotting the distance thrown on passes over careers, and see if this is a good indicator of improvement.
Edit 5: http://phdfootball.blogspot.com/2013/06/field-position-and-s... is interesting, I distinctly remember that Completion % per yardage drops significantly inside 20 yards. I don't have the data on hand, but that's an interesting wrinkle to consider. I think that may have been FO.
I had a similar thought; seems that he should bucket at least by remaining yards to get the first down as well as field position. Imagine an 8-yard pass completion after a 1st and ten: many teams will attempt a deep pass, hoping to pick up the first down with a short play in case of an incompletion. That deep pass is a very different play and probably has a different completion likelihood.
Likewise, consider passes from your own 20 vs their 20 (that is, 80 yards of field left before the end zone vs 20): my intuition says that the latter pass has lower likelihood of completion, because the field is shorter and the defenders have less ground to cover.
So, it might actually be that "rhythm" exists, but only in long field positions or only on long drives.
Disappointed to see this is about "American Football", not "Football".
Seriously, I'm curious about the ever-stricter rules about contrasting uniforms in football outfits - enough that national sides can't play in their classical colors anymore, and club teams have four to six alternate uniform sets; I thought that was to facilitate the automated analysis of games, as I've already seen some contrast-jacked still shots used to overlay tactical diagrams.
As someone else pointed out, there is no one sport that has a monopoly on the term "football". Also, complaining about Americans who refer to American football as "football" often sounds like thinly veiled xenophobia disguised as pedantry, a la "here's one more thing I hate about those stupid Americans..." It's such an odd thing to complain about that it comes across as the tip of a hate iceberg rather than an actual complaint about terminology.
Can you imagine applying this to other countries? "Oh those stupid Chinese, they call their country 'Zhongguo', but it's clearly 'China.'" Seems that labeling country-specific language variations as "wrong" is acceptable only for the USA.
I agree, that part of the post at least seemed on-topic. I guess it was collateral damage, maybe there's a lesson to be learned. (Note, I didn't up- or down-vote)
Alternately, it's like if someone were to write a color management API for Python, but used the British spelling for "color", and a few commenters on some tech forum got up in arms about the spelling. Equally abrasive (the commenters, that is) for similar reasons (a faint whiff of xenophobia and/or a sense of cultural entitlement).
They figured out how to "summarize" video surveillance footage by attaching a hovering timecode to any object moving through the frame and then later it overlays several hours of footage into one video.
Pretty trippy: https://www.youtube.com/watch?v=fISfDd35sXU