With the NFL Draft happening this evening we wanted to provide an overview of what we’re doing with Telemetry Sports in football. Two weeks ago in our co-branded webinar with Telemetry Sports, we explored how we gather football data at Sportlogiq, our partnership with Telemetry Sports, and how our data is used within their platform. The full webinar is available here, but we’ve also taken the time to summarize the key points below.
We focus on game events and tracking (coordinates of the players on the field). To do this we use computer vision to facilitate optical tracking for all 22 players on the field for the duration of the play. Because we do this through the video feeds, there is no need for sensors and therefore no risk of breaking or malfunction.
We track in between the feet, whereas Next Gen Stats tracks players via sensors in the shoulder pads. We use the All-22 feed using the sideline view and endzone view. This lets us see the players’ numbers, we also require a participation list for each game so we can see who actually played.
Applying our computer vision models to these video feeds allows us to generate XY coordinates for every player at 10 times per second (similar to Next Gen Stats).
The tracking itself is done through a combination of computer vision and machine learning. We use a patented process called “camera calibration” to determine the XY coordinates. We do this by overlaying our field template onto the actual game footage for every single video frame of the play. This lets us take all the coordinates (pixel coordinates) and get real-world coordinates for every player on the field for the entirety of the play. It also works on every NCAA field (yes, even Boise State) and every NFL field.
Taking a wide receiver situation for example we can use our tracking data to plot out their targeted route, determine the average separation, their expected yards per target and even what they gained versus what they should have gained on the play.
More than that, we can detect formations based on personnel and previous activity and locations. We can also use the expected yards model to see who is the most advantageous receiving target on a given play. Our data lets us dive even deeper than that and see the instantaneous speed for every player on the field, and offer target ranks to receivers based on expected yardage. Players who are considered “covered” by the defense don’t rank into our expected targets.
In terms of Quarterback analysis, we can analyze if a QB is throwing to the best possible targets. On defense, our likely tackler model is also machine learning-based and takes into account the number of defenders nearby, and where they are headed on a play. This lets us determine what should have happened on a play and the route that should have been taken.
Telemetry Sports is partnered with NFL teams to help them make sense of the data. The most valuable aspect of the data is Telemetry Sports' ability to visualize it. Telemetry Sports visualizes everything in the platform allowing you to see the video of a play alongside the data visualization. You’re also able to look in a grid view and see every play a team ran that season.
Telemetry Sports is also able to analyze all of the data, by ingesting it and creating tags. With that, you’re able to search for and find the most detailed play variations. With a variety of ways to export this information Telemetry Sports really helps coaches determine the keys to the week and simple takeaways that they want to see. Video can also be exported to fit pre-existing analysis systems (XOS Digitial, DVSport). Next-Gen Stats makes this level of analysis possible at the pro level, but it’s the Sportlogiq data that fuels the college product.
This is just the tip of the iceberg of the type of information made available with Sportlogiq and Telemetry Sports.
Follow us on Twitter @Sportlogiq to stay up-to-date with our latest news or contact us for a demo