A Robust Eye Detection Method using Domain Knowledge

H.S. Yoon, D.H. Kim, and J.Y. Lee (Korea)


Eye detection, blob coloring, lip detection


This paper describes an automatic eye detection algorithm using domain knowledge with a low-cost USB PC camera. The object image of this algorithm is the face of a person sitting in a chair or standing in an office. Our algorithm uses no special capturing environment such as lighting, fixed background and pose. Rather, it can detect eye components from facial images under conditions of size variation, complex background, and skew angle variation. The algorithm has two steps. The first step is an eye-based edge verification algorithm that analyzes an isolated region acquired by an adaptive Sobel edge detector. The second step comprises a lip-based edge verification algorithm and, when the input frame has no isolated eye components, a masking-based eye centroid detection algorithm. We tested the proposed algorithm over 2100 frames, which were comprised of 20 facial images each for 105 people. We obtained a 98.7% success rate in the detection of eye components.

Important Links:

Go Back