A Real-Time Eye Detection System Based on Visible Light and Independent Components Analysis

A novel system for real-time eye detection with a hybrid eye detection method of appearance-based and feature-based was proposed. Firstly, the glints can be formed on the eye regions due to corneal reflection characteristics from the visible light. Potential glint regions were obtained based on image algorithm. According to the eye geometrical feature, the candidate eye regions can be extracted well. Secondly, 200 eye images with different appearance were extracted from the eye database. The basis vectors of independent components analysis (ICA) applied on those eye images were estimated based on the FastICA algorithm. Lastly, the accurate eye regions were detected by the projection angles from the candidate eye images to ICA basis vectors. The experimental results showed that the real-time eye detection has a better robustness and high correct rate to different face poses, with glasses, large head movement, and various illuminations.