Use of Context in Automatic Annotation of Sports Videos

The interpretation by a human of a scene in video material is heavily influenced by the context of the scene. As a result, researchers have recently made more use of context in the automation of scene understanding. In the case of a sports video, useful additional context is provided by formal sets of rules of the sport, which can be directly applied to the understanding task. Most work to date has used the context at a single level. However we claim that, by using a multilevel contextual model, erroneous decisions made at a lower can be avoided by the influence of the higher levels. In this work, we explore the use of a multilevel contextual model in understanding tennis videos. We use Hidden Markov models as a framework to incorporate the results of the scene analysis into the contextual model. Preliminary results have shown that the proposed system can successfully recover from errors at the lower levels.

• Type: Conference
• Year: 2004
• Where: CIARP
• Authors: Ilias Kolonias, William J. Christmas, Josef Kittler
• Keywords: CIARP 2004, Multilevel Contextual Model, Scene Understanding, Useful Additional Context