EXTRACTION OF BIOMETRIC INFORMATION VIA VIDEO ANALYSIS USING MACHINE LEARNING : MEASUREMENT OF PALPEBRAL FISSURE HEIGHT AS AN EXAMPLE APPLICATION

Abstract

Methods of human motion tracking for acquiring biometric information are needed in order to apply knowledge of physiological anthropology. However, previous methods have difficulty in control of video shooting conditions. Thus, we developed a more robust motion tracking method using machine learning. In this report, we evaluated the accuracy of our method by comparing palpebral fissure height measured using manually selected data and tracking data obtained using YOLOv3. The results indicate that our method has practical accuracy in measuring palpebral fissure height and suggest that including noise in training data contributes to its accuracy.