The Hudl Algorithm: Turning Video into Player Tracking Data
Hudl is a leading international sports technology company that builds video analysis and stats tools for over 6.4 million pro and amateur coaches and athletes. With the acquisitions of Replay Analysis and Sportscode, Hudl entered the Football market and establish a strong position within the Video Analysis Tools segment. However, this market is seeking more integrated and efficient analysis solutions that offer an end-to-end product for the entire performance analysis stack, which includes capturing video, event data and player tracking data, reports, distribution, and the ability to use tools to do personalized analysis.
Today, teams use different vendors for each of these solutions and string together their own workflows jumping between products. To win with an integrated end-to-end solution, Hudl not only needs to provide a solution in each of these verticals but also strive to be best-in-class. Today, teams are using us for their personalized analysis solution by using Hudl-owned Sportscode, but we needed to expand into the capture, data and reports verticals in order to be a real player in the performance analysis game.
The Goal: Build a scalable player tracking and event tagging system with auto-captured video and analytics to provide a foundation to grow the elite soccer market
With a beta season of player tracking under our belts, we needed to shift our focus to nailing the foundation of the elite football strategy by producing more accurate tracking and event data, and incorporate this data with video in a scalable way.
Our Customers: Elite Soccer Teams
Physical reports from tracking data helps a physical coach determine how a player should recover after a match, what their training should look like every day, and how they should prepare for the next match. A physical coach also warms the players up on the ptich before a match.
Performance analysts use tracking data along with event data to find specific moments in video, such as every time a certain player sprinted in a specific part of the pitch.
Physical reports created from tracking data show how much players ran, at what speeds, and their acceleration levels. This helps a sports scientist determine their nutrition plan after a match, during the week, and after each training session. Sports scientists also monitor a player's sleeping habits and their physiological stats (weight, hydration levels, etc).
How Player Tracking Works
Hudl's player tracking products works by combining auto-capture in-stadium cameras and our machine learning algorithms to track players automatically from video. We then have a team of Hudl analysts who watch the match, make corrections and fill in the gaps where the system may have been incorrect. The player tracking data provides clubs with physical data, which is used to create a range of physical outputs that quantify physical demands on the athlete.
3 cameras are installed in a stadium, and record the game from three angles. We then create a map of the pitch that translates video into an x,y matrix. Once each player is identified with a bounding box, the algorithm runs and records each players movement as x,y data points.
Match Initiation
Lineup Mangement
Before tracking begins, an analyst will use a simple admin page to identify each of the players who participated in the match. This lineup is necessary so that we can link tracking data to the correct players.
Pitch Projection
After a match has been uploaded, a lead analyst will create a projection of the pitch by plotting a set of points on top of the video. We then translate this projection into an x, y matrix on which player movements are mapped.
Tracking Initiation
After the lead analyst has created a projection, they will go through and mark kickoff, optimal tagging time, ending, and extra time. They'll also place a bounding box on each player for our algorithm to track.
Player Tracking
Tracking
After the match has been processed, each player is assigned to different analyst, who ensures they're being tracked correctly. They can move, edit or replace the player marker if it's not placed correctly.
Corrections
After the analyst completed tracking, they would move onto correcting any missing frames or movement errors that were created during tracking.
Tracking QC
Tracking Performance
While a match is being tracked, a lead analyst will observe each job to ensure that each player is being tracked correctly. They'll have the opportunity to correct errors, see if a player is being double tracked, and send a job back to the queue to be re-tracked.
Player Identification
They will also link each tracking job with a specific player from the lineup that they created beforehand.
