Spatio-temporal Modelling using Video Input

Many disciplines study the variation in patterns (events) that occur across space and time. Comparison between these events is often facilitated by the ability to measure the similarity between events based on both spatial and temporal context. This paper describes the initial problems and complexities involved with using spatiotemporal similarity techniques, based upon patterns observed using video input. A template and continuous model of space, incorporating vector representations for movement (time), are both discussed as possible representations for sporting events described by video. The paper includes representation techniques for spatiotemporal events, their comparison using similarity measures and an indexing system for extracting video segments from a database.