A fast method for eye location

Wu Tunhua,Wu Cong,Zhou Changle,Lin Kunhui
2007-01-01
Abstract:Based on AdaBoost algorithm and KLT corner detection algorithm, a novel algorithm for human eye location was proposed. First, cascaded face and eye detectors were trained by AdaBoost algorithm. Second, face regions were detected using the face detector combined with a priori rule of face edge. Third, the rough eye regions were extracted from the face regions using eye detector. Finally, the eye positions were located in the rough eye regions by KLT corner detection algorithm. The experiments on DTU_IMM test set result a correct rate of 99% and a speed of 1 frame per second on 640*480 color images. Results show that the proposed algorithm is efficient.
What problem does this paper attempt to address?