Sports data tracking and analysis
Tracking sport statistics and using the data for performance analysis is nothing new. Cricket and baseball stats were tracked as far back as the late 19 th century. Performance analysis in the form of Sabermetrics has been in use since the early 1970’s. The biggest difference today is the advanced technology and sophisticated data capture and analysis that it enables.
While the methods of gathering data and its uses have evolved beyond what could have been imagined 50 years ago, the core reason for tracking remains the same: creating an edge in performance. This includes analysis of an individual our even specific skills, often with a comparison or benchmarking component, as well as team development such as structural play and player positioning. Increasingly analytics are being used to manage athletes training and workload to help reduce injuries. Media is the other massive consumer of analytic data. It can be used for fan engagement, gaming, and scouting.
What is video tracking
Video tracking is the most widely known technology for gathering sports data. It requires a video recording of a sporting event, which can then later be analyzed through a variety of ways. One of the most common is post game analysis, where coaches and players review the recorded game, or clips of specific plays or game events. Another common method is to have third parties manually tag events within the recorded session, thereby generating a list of events that corresponds to the video timeline. Getting more sophisticated, machine learning can be trained to track players and puck or ball movement. This is known as computer vision. To be accurate this requires good video that captures all aspects of the game unobstructed, and training the algorithm with specific video angles.
What is sensor-based tracking, and different types of sensors
With sensor-based tracking a sensor or array of sensors are used to gather data on players and puck or ball movement. This could be a motion sensor that records acceleration and change of direction, such as the step counter in a Fitbit, or a GPS tracker, some technologies use the GPS in your phone, or a series of sensors which track a tag or beacon within the field of the sensor array. As with the video technology, machine learning is used to parse and analyse the data gathered to offer meaningful reports. The motion sensor and GPS tracking are what is known single sensor systems. The Drive Hockey Player and Puck Tracking System is built on an array of sensors around the rink with tracking sensors on the players and in the puck. This is known as a multi-sensor system and we’ve chosen to develop this type of sensor tracking for reasons that will become clearer when we get to comparing systems.
The role of artificial intelligence
Increasingly AI is used to handle data rich analysis. For systems such as ours which produces over 3000 data points a second, it is critical to take this huge amount of data and convert them to manageable metrics. Machine learning algorithms are trained on data sets to be able to recognize meaningful patterns and convert the millions of data points into simple to understand metrics. The important thing to understand is that each algorithm is trained with different types of data and built for analysis of specific outputs, which means that the quality of the output depends on the input data. If the input data is poor quality or in limited quantity, then even the best trained algorithm will produce poor outputs.
Comparing sports tracking systems – advantages and drawbacks
| DRIVE HOCKEY | SINGLE PLAYER SENSORS | VIDEO-BASED | |
|---|---|---|---|
| OPERATION | Simple, setup and takedown in 20min., automated tracking | Simple, self-provisioned, automated tracking | Manual video capture & event tagging |
| PROCESSING TIME | Real time | Real time | Up to 1 week |
| MARGIN-OF-ERROR | < 0.2m | 1.5 – 5m | ~1.5m |
| DATA POINTS | 10 – 20 /sec. | 5 – 10 /sec. | Once per event tracked |
| DATA RICHNESS | 3000+ metrics collected per second | 50+ metrics collected per second | 5+ metrics collected per second |
| AI / ML | AI models process data, identify events, skills and techniques | AI models process data | AI models help identify players and positions |
| SCOPE OF ANALYTICS | Athletics and gameplay at player and team level | Some athletics at player level | Athletics and gameplay at player and team level |
| ANALYTICS QUALITY | Deep intelligence of skills, events, outcomes, and context | Some insight of individual player performance | Some high-level stats as a secondary feature to game film breakdown |
| OBJECTIVITY | Fully objective | Fully objective | Requires human validation introducing subjectivity |
| EVALUATION | Instant assessment of player skills, tactical play, impact and decision making filtered by situational play. | High-level single data points around athletics | Video clips categorized by events. Requires time to review individual clips. |
| SKILL DEVELOPMENT | AI-driven skill analysis and breakdowns, how to improve each, track progress over time and benchmarking. | Track progress over time and benchmarking | Requires time & expertise to identify and evaluate player techniques vs desired outcomes. |
| COACHING | Real-time data access, practice & game breakdown, play review, line combos, situations analyzed and consolidated reports | Consolidated reports | Game breakdown, play review |
Data at time of publication is based on Drive Hockey vs sampled competing technologies
One of the major advantages with video is its ability to provide great context in post game analysis, especially when paired with marked game events. Video however has several drawbacks, primarily the line-of-sight aspect which means you only capture what you can see, in addition to the need of high-quality video of the event in order to process meaningful data. As such, many video-based platforms can become cost prohibitive or are not appropriate for amateur levels. Another challenge for video is that
many systems rely on human verification of the data, adding cost and time to production of the analytics as well as introducing subjectivity and inconsistency to the performance analysis process. Often, insights are not available for days after the event has taken place. Although video recording in public is lawful for personal use, if the same recording is used for commercial purposes, it requires consent from those being recorded, which becomes challenging and sensitive at the youth levels.
Looking at sensor-based sports analytics systems, there is greater accessibility for amateur teams and players, as well as for use during practices because it doesn’t require a quality video feed. Sensor-based systems also allow for real time data since there is no human element, and they are generally all opt-in systems by design. Sensor-based systems can also capture data that the eye does not see – it will capture data regardless of obstructed views and capture new data types limited only by sensor technology.
Another access consideration is cost. Single-sensor technology tends to be the winner here because it has the least hardware. Some systems piggyback on the GPS in your phone. The drawback is single-sensor systems are typically the narrowest in scope when it comes to what can be measures and how it relates to the overall game. These systems provide individual performance data but not what else is happening on the ice.
Perhaps the biggest differences are in terms of the data collected, and therefore the potential of the analytics. Computer vision relies on people to identify a very general event, such as a shot, but does not provide accurate data around the situation, technique, speed or result of the shot. Event data produced from video may result in as few as one data point every couple of seconds and offer location accuracy of about 1.5 meters. In perspective, the puck is 7.5 cm (3 inches), or 20 times smaller than the accuracy range. Single-sensor systems can collect 100’s of data points per second but tend to have a location accuracy range of 1.5-5 meters, making it difficult to use for meaningful data. Single sensors can also be limited in scope, providing only one-dimension of data, such as acceleration of an athlete, whereby other data points are extrapolated without a high degree of confidence.
Data richness and accuracy is where the multi-sensor systems excel. Drive Hockey built its technology using the latest location tracking technology, similar to what the NHL is also using in their systems. Unlike the NHL system, our system is engineered to be mobile and able to setup and take down quickly before / after events, and has a price tag that amateur hockey can afford. Whereas GPS based systems fail in terms of accuracy, ours produces location accuracy of up to 20 cm simultaneously from multiple players and objects and at resolutions in the thousands of data points per second. These location measurements, combined with other sensors, provide our AI algorithms with a complete athletic performance assessment, such as detailed skating maneuvers, puck possession, battles, passes and shots. In addition, since there are sensors all around the rink, on each player and in the puck, it is able to capture a complete picture of the game. This means that we can also compare and analyze team performance, tactical decisions and game play in relation to the opposition.
Why the trend towards better data?
Typically, when hockey people think analytics, they think of things like Corsi, shot-attempts and the fancy graphs published in media. This is typical outcomes of event-driven data, such as produced through video analysis. Although sensor-based systems can provide this same data as well, they provide another richer layer of intelligence. Whereas an experienced scout, coach or trainer may need to sit down and review hours of video to assess a player’s strengths and weaknesses, sensors provide an
immediate deep assessment of the same players skills in measurable, easy to understand ways. Player’s skills can be compared and benchmarked against their peers, or even to higher competition levels providing clarity on exactly what is needed to advance their game. The future of sports data tracking and analytics Ultimately sport data tracking and analytics are here to stay and we are at the tip of the iceberg in term of realizing the full potential. In the foreseeable future, at the professional level, we expect to see the use of a range of technologies. This is most likely a combination of multi-sensor and computer vision systems that work together. Uses will include tracking and monitoring individual performances and athletic data, benchmarking, team performance analysis and post game analysis, and more. The data will also increasingly be used by media, gaming and scouting. Just think about your next season of fantasy hockey with real time data, betting, replays and player performance benchmarking.
Hopefully this provides some clarity on the different types of tracking and analytics systems. If you’re interested in learning more about Drive Hockey’s Player and Puck Tracking System for your team or league or want to be a partner that brings it to your region please contact us.