By Fahong Li
Motion is an important cue to the intentions of active agents in environments involving collaboration and competition. We demonstrate this in the domain of ice hockey. We develop a framework to represent and reason about hockey behaviors using as input actual player motion trajectory data tracked from game video and supported by knowledge of hockey strategy, game context and specific player profiles. The raw player motion trajectory data consists of space-time point sequences of forward/backward skating registered to rink coordinates. This is augmented with knowledge of possession of the puck and specific player attributes (e.g., shoots left, shoots right).
We focus on the analysis of three clearly identifiable situations: 2-on-1 offensive attacks, defensive zone breakouts and power play shots from the point. We use a Finite State Machine (FSM) model to represent our total knowledge of a given situation and develop evaluation functions for primitive hockey behaviors (e.g., pass, shot). Based on the augmented trajectory data, the FSMs and the evaluation functions, we describe what happened in each identified situation, assess the outcome, estimate when and where key play choices were made, and attempt to predict whether better alternatives were available to achieve understood goals. A textual natural language description and a simple 2D graphic animation of the analysis are produced as the output. The graphic animation is useful for interactive visualization and debugging. The textual description also provides potentially useful annotation for large databases of player motion trajectories.
The framework is flexible to allow the substitution of different analysis modules and extensible to allow the inclusion of additional hockey situations. We expect that the methodology and the framework can be generalized and applied in other domains, such as soccer, basketball, traffic flow control and people surveillance.
Back to the LCI Forum page