Soccer video analysis from your match videos (futvis.com)

Wow! This is cool! Have you seen the video where they use OpenAI to analyze and give commentary on live soccer games by sending screenshots of the game to OpenAI vision API and text to speech API?

https://youtu.be/u56K4dL20gA?si=VBtBicEQ89f1l6Fv

Cool. I've been thinking about where to find an alternative to the Veo Technologies camera system that most clubs use but is pretty expensive. This seems like it could work

would this work for other sports like rugby?

Hey, this looks pretty cool. I play with a couple of teams that are pretty competitive in local indoor leagues. We would like to try this out. I applied for the beta, but wonder if there is a way to contact you directly. Some questions:

- Approximately how much does it cost ?

- Can it use videos of the same game, shot from different angles ?

This is awesome. Would love access for my daughter's soccer team.

This is great, and I hope people enjoy it, I tried to do something similar years ago commercially, and so I'll ask: please make sure you keep it as a passion project - turning this into a money spinner could lead to disappointment.

I was on the winning team for what I believe was the EPL's first and only hackathon at Manchester City back in 2016 or thereabouts. It was a while back.

That hackathon used a couple of data sources that professional football (never "soccer" on this island), has available: Opta data which is now the industry standard, and a 25fps video system turned into player, official and ball tracking data based on 6+ cameras installed at every ground in the league. Teams had access to all games they participated in, but not all their competitors games. Hackathon brief: what you can do with these two data sources?

That video tracking at 25fps was not great. Yes, producing heat maps, cool. Yes, automated pass tracking, cool. Automated distance travelled, pace/minutes played, possession, yada, yada, all cool. Combining it with Opta data meant you could start to get pass completion and "forced error"-like stats, cool. So you could start to think about how to get outputs that could turn into coaching inputs, and that's kinda cool but also, y'know, coaches like Pep aren't going to listen, no matter how often you make them read/watch Moneyball.

We won because we quantified the effect of opposing players on a 30 degree and 45 degree forward arc on pass completion - intuitively any fan knows that defending players lower pass completion, but we were the first in the World apparently to put a hard number on it because of this data. (12% reduction in pass completion for every player in a 30 degree arc ahead of a player, if you're interested).

However, we had suspicions with video based analysis for accurate tracking. Because of - when you think about it, kinda obvious - timing issues that can mean at 25fps you have 40ms latency/slippage frame to frame, the data suggested the ball was kicked in one game, by Yaya Toure into the back of the opposing team's net at just over 1000m/s - a little over Mach 3. I remembered the goal from that game and quipped "it was good, but not that good".

After doing that work CFG started to explore next steps. That's a long story, but I started to dive in to some ideas. I pointed out that there was so much archive video from TV coverage that could be used for analysis, throwing it all through some pose estimation could yield results. Even just set pieces, even just corners, could mean you could produce a huge resource for analysis.

The problem is, there's not much budget for this.

It seems odd, but most EPL clubs - and EPL is the richest league in the World - have annual turnovers that would be dwarfed by some Series D startups. All the money goes on the field in player salaries. If you've read Soccernomics, you'll agree it should. That meant data teams often have small budgets, measured in the high 5-figure/low 6-figure region for kit, and not much more for analyst salaries.

Digging around, I discovered that there are high school grid iron football teams in Texas whose non-salary OPEX budgets for data analysis aren't that much lower than most EPL teams.

That may have changed, but I think you have a chance of making a dent in this space, getting interest from lots of clubs (both professional and amateur), but you might struggle to make decent returns from it. It's why I had the exact same idea years ago and abandoned it - it seemed like a lot of work for very little return and I needed to focus on returns at that time.

TLDR: this is awesome, I hope it thrives, keep going, and Godspeed. Just keep the day job for now. ;-)

Really cool! Was thinking to build something similar myself as a side project. What is the tech stack? I would be curious to learn more about how it works.

If someone wanted to make something like this for tennis I have several large academies that would love to use it (could easily get a couple co-founders like the UTR founder to offer it to the tournament circuits and associations and make it a robust business)

Does this work in real time? Considered gambling applications?

There is a decent market for these kinds of tools if you can make easy for users to use. Back in the late 90s/early 00s, I worked for a company that provided MPEG2 equipment to startup making these kinds of analysis tools for American Football analysis. The NFL has had it for years, but they were trying to make it affordable for High Schools to get in on the action. I'm in Texas, so that's pretty much a religion, so you can bet they spent the money.

This was very manually oriented where the students would capture the VHS tapes schools swapped with each other, and then log each play, the down, the time on the clock, and every other piece of data they could think of. At the end, it would analyze it all and be able to show that the coach from School A has tendencies to run a particular play or defense in the 4th quarter. It was pretty accurate.

If you can do the same thing but without needing the manual data entry, there could be a product instead of just being a fun project. If you can show that certain players have more tendencies to switch to the left foot after cutting in from the right, the defense can look for it and shut down that lane and offer the outside (isn't that their default anyways???) to make the attacker chose their "less preferred" action. We know goal keepers study PK attempts in attempts to have a more informed guess on the shooter's tendency on placement.

Never underestimate the money schools/coaches will go to gain an advantage. Hell, we're seeing yet again coaches in trouble for trying to steal the play calls. The one positive about trying to sell it is that you do not have to worry about the school's budget (which is never enough) since this is the exact kind of thing the boosters love to fund in an attempt to get the upper hand on a rival.

Good luck!

There are a couple of other products in the market doing very similar things. Have you done any work on positioning or differentiation? There is space for different options for sure.

I'd love to see if you can turn this into an affordable virtual offside system for smaller leagues.

This is not a valid Show HN - you have to have something more than a signup list for people to try out. Please see https://news.ycombinator.com/showhn.html.

I appreciate that you have a demo of the project working on one video, but the title claims that people can try it on their own, and for that the web page seems to just be collecting signups. Most likely it would be best to do a Show HN once it's working in a form people can try directly.

(I've taken "Show HN" out of the title now.)

Questions: - Can you do analysis of video captured from Trace cameras? - Any roadmap for exporting the data for both team and player?