A new appearance based and user independent eye state detection method using eigeneyes

Eye state detection is the process of determining whether the eyes are opened or closed. It has been widely employed in driver fatigue/drowsiness detection, human computer interaction. However, there are still lots of problems remaining unsolved such as free head movements, changing pose and illumination. Due to this, a new appearance based method is investigated in this work. Appearance based eigeneye features are extracted using Principal Component Analysis subspace method which is also a dimension and noise reduction method. And, for the first time in literature, eigeneyes are used for eye state detection in this work. K-nearest neighbor and multi-layer Perceptron Neural Networks with back-propagation learning algorithm are adopted to estimate eye state. Eigeneyes and Perceptron Neural Networks method pair obtained user-independent accuracy of 92.13%. Compared to previous works, the following two points are improved: (i) a new appearance based feature extraction method is proposed for eye state detection, and (ii) a faster approach is obtained for real-time systems while preserving accuracy substantially.