Multiple Person and Speaker Activity Tracking with a Particle Filter

Neal Checka, Kevin W. Wilson, Michael R. Siracusa, and Trevor Darrell

This sequence consists of 1836 frames and shows one to three people walking around the room and conversing. In this experiment, the particle filter was run with 200 particles. The rectangles represent the state of the world as determined by our estimation algorithm. The darker rectangles denote active speakers, while the lighter rectangles represent non-speakers.

Demo