Robust visual voice activity detection using chaos theory under illumination varying environment
Taeyup Song, Kyungsun Lee, Hanseok Ko · 2014
We propose a visual based voice activity detection robust in illumination changes which frequently occurs in indoor vehicle environment. By applying chaos theory, contextual change becomes chaos-like and results in complex fractal trajectories in phase space. We measure fractality in phase space from image frames and thereby robustly detect visual speech motions under illumination changes. Representative experiments demonstrate significant improvements over conventional method.