This paper presents the audio based tracking system designed at FBK-irst laboratories for the CLEAR 2007 evaluation campaign. The tracker relies on the Global Coherence Field theory that has proved to efficiently deal with the foreseen scenarios. Particular emphasis is given to the post-processing of localization hypotheses which guarantees smooth speaker trajectories and is crucial for the overall performance of the system. The system is also equipped with a speech activity detector based on Hidden Markov Models. The performance delivered by the proposed tracker presents a considerable gain with respect to the previous evaluation. An attempt to devise a multimodal tracker based on merging outputs of a video and an audio trackers is also described.
A Person Tracking System for CHIL meetings
Brutti, Alessio
2007-01-01
Abstract
This paper presents the audio based tracking system designed at FBK-irst laboratories for the CLEAR 2007 evaluation campaign. The tracker relies on the Global Coherence Field theory that has proved to efficiently deal with the foreseen scenarios. Particular emphasis is given to the post-processing of localization hypotheses which guarantees smooth speaker trajectories and is crucial for the overall performance of the system. The system is also equipped with a speech activity detector based on Hidden Markov Models. The performance delivered by the proposed tracker presents a considerable gain with respect to the previous evaluation. An attempt to devise a multimodal tracker based on merging outputs of a video and an audio trackers is also described.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
