The general aim of the Computational Analysis of Ice Hockey project is to develop a system capable of 1) learning the way ice hockey is played and 2) using this knowledge to enhance team performance. More specifically, we are interested in analyzing video from a particular hockey team, automatically detecting play styles, propensities, and habits and then using this information to suggest strategies for improving a team's performance. Our work draws on ideas from machine learning and machine vision.
Students: Derek Kwok, Mansoor Siddiqui Collaborators: David Fleet (Dept. of Computer Science)
We are currently working on proof-of-concept algorithms for:
Camera Tracking: A moving video camera observes different locations of the ice at different times. Building an accurate camera model by modeling the camera's intrinsic and extrinsic parameters allows us to know where we are looking at any point in time.
Player Identification and Tracking: To understand play styles, we need to automatically determine, from the video sequence, where players are located on the ice (player identification) and subsequently how they are moving (player tracking). Player identification and tracking is made difficult by visual obstructions.
Puck Tracking: The transient nature of puck visualization and the speed with which the puck moves makes puck localization extremely difficult.
Reasoning Under Uncertainly: Algorithms capable of reasoning under uncertainty are ubiquitous in scientific computing. Building from our experience in Computational Biology, we are reasoning about play styles without complete knowledge of game state.
Learn more about each project by clicking through to their project pages.