Most eye detection methods suffer from the influence of illumination and face pose. This paper presents an integrated eye detection algorithm to overcome these limitations. First, we have designed an effective illumination normalization method to overcome variable lighting condition, then use a pose independent Adaboost method to detect faces, for any detected face image, we present a feature point extraction method to extract face feature candidates. In the end, a heuristic rule is used to filter non-eyepair candidates, and the Support Vector Machine (SVM) are employed for verify eye-pair. Experiment results indicate our algorithm can achieves excellent performance under variable illumination and various face poses conditions.