Eye Detection and Tracking Using Rectangle Features and Integrated Eye Tracker by Web Camera

Eye detection and tracking can be used in intelligent human-computer interfaces, driver drowsiness detection, security, and biology systems. In this paper a new method for eye detection based on some new rectangle features is proposed, with these features the Adaboost cascade classifiers are trained for eye detection. Then with the characteristics of symmetry of the eyes some of the geometric characteristics are adopted for correction. The geometric characteristics improve the accuracy of the eye detection, and make the rough cascade classifier trained by few samples become a reality in application. In this paper, we present an integrated eye tracker to overcome the effect of eye closure and external illumination by combining Kalman filter with Mean Shift algorithm. Results from an extensive experiment show a significant improvement of our technique over existing eye tracking techniques. According to the taxonomy proposed in (4), the techniques for eye detection and tracking can be classified as shape-based (5, 6), feature-based (7), appearance-based (7-11), and hybrid (12), based on their geometric and photometric properties. Shape-based methods can be classified as fixed shape and deformable shape. While the shape-based methods use a prior model of eye shape and surrounding structures, the appearance-based methods rely on models built directly on the appearance of the eye region. Hybrid methods combine feature, shape, and appearance approaches to exploit their respective benefits. For example, an open eye is well described by its shape, which includes the iris and pupil contours and the exterior shape of the eye (eyelids). Shape models are usually composed of a geometric eye model and a similarity measure. The parameters of the geometric model define the allowable template deformations and contain parameters for rigid transformations and parameters for non-rigid template deformations. In this paper, a new method for eye detection based on rectangle features and geometric characteristics is proposed. Rectangle features were firstly proposed by Paul Viola, in Paul Viola's paper integral image and cascade classifier were proposed and these methods made the detection system real time. This paper uses Viola's conception, and adds some new rectangle features to construct a cascade classifier for rough eye detection. The results from the classifier usually have some errors, such as eyebrows, mouth, nares, larger or smaller eyes, so geometric features are introduced to assistant to detect the eye location accurately. And then, an integrated eye tracker to overcome the effect of eye closure and external illumination by combining Kalman filter with Mean Shift algorithm is presented. This eye tracker can robustly track eyes under variable and realistic lighting conditions and under various face