Muithu – A sports notational analysis system

Digital information systems might provide the comparative benefit that distinguishes top athlete performance from their competitors. For instance, video footage of sport athlete performance is widely used to provide accurate and objective information to top athletes and their coaches.

Notational analysis

Digital notational analysis systems can be used complementary by providing a list of measurable parameters based on the observed performance. Such quantitative parameters are relative to the sport and activity at focus, and example events include that a specific soccer player passes the ball into the opponent end-zone, or that a rugby player is being tackled. Some notational analysis systems require manual annotation of such events while others are more semi-automated. We are in particular interested in sports with non-linear flowing performance, which makes it harder for automated analysis. Soccer is an example of a flowing game, and many premier league soccer clubs already use this type of systems.  This includes Barcelona FC, Inter Milan, New York Red Bulls, Manchester United, Liverpool FC, Chelsea FC, and Arsenal FC.

Unfortunately, there are limitations and problems with several of these notational analysis systems. For instance, humans with domain expertise still need to manually identify performance events of a certain complexity. Head coaches can best determine perceived performance against complex, yet often detailed and dynamic strategic goals, but rarely have capacity for this tedious and time-consuming analytics process.

A novel light-weight approach

As a compromise, a collection of static cameras are wired together with computers running analytics software, but where a set of additional human experts also needs to be in the critical analytics path. With such a large footprint and operational cost, applicability of existing systems is primarily for main events like games, not for day-to-day training sessions and practices. We have developed Muithu, a complementary, light-weight video and cellular phone based notational analysis system for this purpose. Probably contrary to what to expect for a small footprint system, have we added the head coaches to the notational analysis loop. Our main conjectures for such a digital information system include that it has to be non-invasive for the head coaches, yet independent of a large group of human analysts in the back-end. The system also has to be close to fully automated, mobile, scalable, with a minimalistic footprint and low operational cost, and, last but not least, applicable in close to real-time. This set of properties is, if we may say so, novel and might have a large imprint if possible to retrofit them all into a single, applicable system.

Secure Abstractions with Codecaps

Obviously, there are strong security constraints related to athlete and team performance data.  For example, medical related information like heart-rate and injuries are highly personal and cannot be made public. Muithu is designed to observe security requirements from the ground up.

One can think of Muithu as consisting of layers of abstraction.  Each layer implements its own services and supports operations through a remote procedure call mechanism.  Access to data is mediated through codecaps.  Services are run by principals; clients that access services are principals as well.

The base-layer of Muithu consist of captured notational data, video feeds, and sensor data that are pushed to and stored on an enterprise
cloud platform.  This set of data, hosted by the base-layer principal, is represented as a set of data objects that can be accessed through a simple interface.  Such data objects might, for instance, correspond to raw sensor data of individual players in the team, and might be updated as new data about that player becomes available. Additional layers are then added as the data is being processed and tagged.  Some layers have significant cloud resources available, but others work more like a library executed by their clients, often using JavaScript in the browser.  The cloud resources of such layers are only accessed when the library cannot handle requests itself.