MIT Lincoln Laboratory multimodal person identification system in the CLEAR 2007 Evaluation
May 8, 2007
Conference Paper
Author:
Published in:
2nd Annual Classification of Event Activities and Relationships/Rich Transcription Evaluations, 8-11 May 2008, pp. 240-247.
R&D Area:
R&D Group:
Summary
A description of the MIT Lincoln Laboratory system used in the person identification task of the recent CLEAR 2007 Evaluation is documented in this paper. This task is broken into audio, visual, and multimodal subtasks. The audio identification system utilizes both a GMM and a SVM subsystem, while the visual (face) identification system utilizes an appearance-based [Kernel] approach for identification. The audio channels, originating from a microphone array, were preprocessed with beamforming and noise preprocessing.