Lip Detection in Video using AdaBoost and Kalman Filtering

Abstract

Lip reading is an active field that receives much attention from computer scientists. Its applications take part not only in science, such as a speech recognition system, but also in social activities, such as teaching pronunciation for deaf children in order to recover their speaking ability. In this paper, we aim to solve a narrower problem, the lip tracking, which is an essential step toprovide visual lip data for the lip-reading system. Inspired by the idea of AVCSR, which has combined visual features with audio features to increase the accuracy in noisy environments, we use AdaBoost algorithm and Kalmanfilter for the face and lip detectors. Our result shows that the system can detect and track mouth motion in real time. It is a critical point that encourages more researches in the visual tracking and voice recognition fields.