A Generative Approach to Audio-Visual Person Tracking

This paper focuses on the integration of acoustic and visualinformation for people tracking. The system presented relies on aprobabilistic framework within which information from multiple sourcesis integrated at an intermediate stage. An advantage of the methodproposed is that of using a generative approach which supports easyand robust integration of multi source information by means of sampledprojection instead of triangulation. The system described has beendeveloped in the EU funded CHIL Project researchactivities. Experimental results from the CLEAR evaluation workshopare reported.