Image Processing Methods For Mobile Horizontal Gaze Nystagmus Sobriety Check

A method for eye gaze measurement is presented for the intended purpose of alcohol induced Horizontal Gaze Nystagmus detection. The method uses two pupil locating techniques on two still images, followed by geometric estimation to perform initial facial feature estimation. Laplacian-of-Gaussian filtering in scale-space is used to locate the pupil center on an image where the eye is centered, and gradient techniques are used to locate the pupil when the eye is not centered. After the initial geometric estimation, template matching techniques are used for video frame registration, and integral projection techniques combined with one dimensional template matching are used for fast, real-time gaze measurement. The method is designed to require minimal computation on video frames, and is tailored for use on a mobile device platform. A test implementation on a PC, using prerecorded video data, successfully tracks a noisy sinusoidal eye gaze with a mean error of approximately 11 percent.