Over the past 10 years, the method in which sports are watched, played, coached, officiated, broadcasted and organized have been greatly transformed by new technologies.  Computer vision has played a central role in this transformation as the majority of new technologies have been vision-based,  for example: vision-based systems such as Hawk-Eye have provided ball tracking for enhanced broadcast visualizations and due to their accuracy have also been used for aiding umpiring decision in both tennis and cricket; virtual insertions, like the first-down line in American Football or world record lines in athletics or swimming have increased viewer excitement; real-time pitch-tracking in baseball showing the strike-zone has improved the level of analysis; and more recently, vision-based systems have been used as a solution to detect when a goal is scored in soccer to avoid recent goal-scoring controversies.

Even though tremendous progress has been made, there are still many problems to solve. Most notable are those of  fully automatic player and ball tracking in continuous team sports; activity recognition (e.g. automatic statistic generation); team tactic analysis and prediction; marker-less motion capture of athletes and bio-mechanical analysis; and automatic broadcast solutions. As nearly all professional leagues currently forbid the use of wearable sensors on players, the unobtrusive nature of computer vision makes it the preferred approach.

In the vision community, a substantial amount of research has been targeted in these unsolved areas, but a major bottleneck has been the lack of a large dataset which is available for researchers to access and compare approaches on. This is due to the fact that capturing, storing, annotating and distributing such data is very costly and time-consuming, which makes it difficult for academic institutions to collect such data, while due to the appealing commercial applications, collection efforts within the industry domain have remained private for proprietary reasons. However, due to the complexity and challenge of these problem, it has come to the critical point where both academia and industry needs to work together in an effort to solve some of these issues.

The goal of this workshop is two-fold: 1) release a dataset of a complete basketball game to the research community, complete with manual player tracking annotations, identities, labeled activities, complete with a baseline approach and protocol, and 2) bring together top researchers in academia and industry together to talk about these problems and foster potential collaborations.  As the largest available dataset is only two minutes, our dataset of close to 60 minutes represents a significant contribution to the vision community.

The workshop will depart from the traditional approach of having paper submissions, and will be confined to invited speakers.  We plan to have 10-12 invited speakers, with a split between academia and industry. Each speaker will be allotted 30 minutes each. At the end of the workshop, we will be announcing the release of the database to the research community.

List of confirmed speakers:

Irfan Essa (Georgia Tech)

Bob Collins (PSU)

Jim Little (UBC)

Yaser Sheikh (CMU)

Chris Bregler (NYU)

Rajiv Maheswaran (USC)

Christophe De Vleeschouwer (UCL, Belgium)

David Martin (AIS - Cycling)

Peter Blanch (Cricket Australia)

David Rath (Hawthorn Football Club)

Stuart Taggart (Relevant Innovation)

Gus Seebeck (ESPN Australia)