StreamTeam: From Individual Sensing to Collaborative Action Analysis (Ongoing)

logo

With the recent proliferation of small, unobtrusive embedded sensors, the number of data streams and the volume of streamed data has increased enormously. This has strongly influenced both our business and our private life and has brought forward a large variety of monitoring applications in different domains. In all these applications, the analysis of data streams in real-time is essential. One of the main challenges in data stream analysis is the detection of complex events out of the raw streaming data. While personal assistants allow to analyze in detail the activities of individuals, the analysis of entire teams, in particular complex events reflecting the interaction and relations between team members on the basis of the data streams captured on the individuals is largely unexplored. At the same time, such team analysis is essential in all applications where groups of individuals have to jointly solve complex tasks. Examples can be found, for instance, in team sports or in rescue operations (fire fighting, disaster management, etc.).

The goal of StreamTeam is to monitor, analyze, and visualize the actions of teams of individuals in real-time in highly dynamic and mobile environments in a robust and scalable way. StreamTeam meets the challenges arising from the very rapidly increasing volumes of continuous data streams that will be digitally captured on the individual and goes beyond mere data collection about sensed facts and log-based post-hoc analysis of the actions of individuals. Most importantly, StreamTeam takes a major step towards online inference of group behavior.

So far, StreamTeam provides i.) novel algorithms for the detection of semantic events in streams of data coming from highly mobile sensors, ii.) a scalable platform for the management of data streams and the detection of semantic events in real-time, and iii.) a modular approach towards specifying analyses (incl. complex events and high-level strategies) thereby bridging the gap between a generic platform and selected analyses in a concrete application.

As a next step, StreamTeam will be extended towards supporting more sports disciplines. Moreover, we will integrate deep learning approaches to detect complex events for which an algorithmic specification is not available. For this purpose, we will investigate if deep neural networks can be trained on the basis of existing (annotated) data sets. In addition, we will investigate if coach-specific data sets that are deduced from explicit or implicit event annotations in SportSense, our team sport video scene retrieval system, can be used to train coach-specific event detection networks.

The StreamTeam project will continue focussing on team sports, and in particular on football, based on the existing collaboration between the DBIS group at UNIBAS and the BFH Centre for Technologies in Sports and Medicine. As the Swiss Federal Institute of Sport Magglingen (SFISM) is part of the BFH centre, the latter acts as interface between technology and especially computer science and sports. The BFH centre closely collaborates with the Swiss Football Association (SFV) and in particular with the coaches of the youth national teams of the SFV. In addition, the BFH centre has also a close collaboration with the International Ice Hockey Federation (IIHF). As this community is currently in the process of preparing the basis for deploying sensors in the equipment of the individual players, ice hockey will be considered as a second use case for the evaluation of the project results as soon as these technical developments further proceed.

A recent video illustrates the collaboration between the DBIS group and the BFH Centre for Technologies in Sports and Medicine / the Swiss Federal Insitute of Sports (SFISM):

 

A video showing the real-time user interface of StreamTeam is published on Youtube:

 

In December 2017, an article on StreamTeam was published in Tageswoche.

Moreover, the whole code of StreamTeam is published on GitHub under the GNU Affero General Public License v3.0.

Since

01.08.2015

Partners

BFH Centre for Technologies in Sports and Medicine (Martin Rumo)

Funding Agencies

Funded since September 2017 by the Hasler Foundation in the Cyber-Human Sytems program (contract no. 16074).

Staff

Research Topics

Publications

2020
2019
2018
2017
2016