An Efficient Recognition Method for Drivers' Eye States

摘要

In order to decrease the influence on the recognition of for drivers' eye states when the lightness or view angle change suddenly, a new algorithm in this study is proposed to improve the recognition rate, which combines Kanade-Lucas (K-L) optical algorithm with Adaboost cascade classifier. This algorithm recognizes and saves the Harris corner using Adaboost algorithm. Those saved corner features would be tracked using K-L algorithm, if Adaboost algorithm did not recognize them again. The method improves the recognition rate and reduces the iterate computation of identification. Because, it is difficult to distinguish the eye with the eyebrow or the eye-rim, the second threshold segmentation algorithm is advanced to decrease the initial value of the global threshold and the second threshold is set to improve the recognition rate. Freeman chain code is used to search the contour of the recognized eye and three states of the drivers' eye are determined according to Elliptic equations. The experiments showed that the method can decrease the influence of the lightness and view angle and improve the recognition precision efficiently.