Driver face tracking using Gaussian mixture model(GMM)

For the purpose of driver fatigue or attention level surveillance based on various facial cues, a practical computer vision system would probably implement at least three major components: a face detection module to locate the face as system initialization, a face tracking module to track the face in the subsequent images, and an inference module to inference the driver vigilant status from certain facial features. Each of them is of research interest as an independent component. In this paper, we describe a driver face tracking method using a semi-parameter probability density estimation method. Assuming that an initial face location is by a certain reliable face location module, our tracking method starts with an estimation of a two-component Gaussian Mixture Model as the initial face pixel gray value distribution using an iterative EM algorithm. Then, the estimated Gaussian Mixture is used to locate the face in the next image frame. Also this Mixture model is updated by a recursive EM algorithm using pixel gray values from the tracked face, and the updated model is to be used to locate the face in the next image. We illustrate this method through experiments on the video sequence captured by IR-sensitive camera.