CAJ | 학술논문

针对目前人机交互中存在的问题与不足，提出了一种基于多模态视觉特征的人机交互新方法。采用二维Gabor小波提取全局人脸朝向视觉特征，克服人脸显著特征提取困难，以及不同人脸朝向特征区分的难度性；基于人脸几何分布特征，快速定位人眼中心，无需考虑人脸图像分辨率和人眼闭合及其配戴情况；通过对视觉特征的重要性评价，选取分类特性显著的多模态视觉特征进行机器学习与训练，确定用户所指目标，实现非穿戴自然人机交互，用户无需佩戴任何标记，且其活动不受约束，便于充分发挥其日常技能。通过实验对比，验证了文中所提方法有效、可行，可应用于实时非穿戴自然人机交互中。
침대목전인궤교호중존재적문제여불족，제출료일충기우다모태시각특정적인궤교호신방법。채용이유Gabor소파제취전국인검조향시각특정，극복인검현저특정제취곤난，이급불동인검조향특정구분적난도성；기우인검궤하분포특정，쾌속정위인안중심，무수고필인검도상분변솔화인안폐합급기배대정황；통과대시각특정적중요성평개，선취분류특성현저적다모태시각특정진행궤기학습여훈련，학정용호소지목표，실현비천대자연인궤교호，용호무수패대임하표기，차기활동불수약속，편우충분발휘기일상기능。통과실험대비，험증료문중소제방법유효、가행，가응용우실시비천대자연인궤교호중。
A novel human-computer interaction(HCI)is developed based on multimodal visual features aiming at some limits at present .Two-dimensional Gabor wavelet is adopted to extract some visual features of global face orientation ,which overcomes some difficulties including extraction of some facial distinct features ,discrimination among some different facial orientations .An ef-ficient and fast approach to locating center of eyes is proposed based on facial geometric distributions without considering facial res-olution ,eyes closing or opening and user′s wearing .Some prominent multimodal visual features for classification are selected to ma-chine learning and training to determine the pointing target after evaluating performance of some extracted visual features .Non-wearable and natural HCI modal can be realized in which user can move freely without wearing any markers when he points at some targets .Their daily skills can be exerted fully during HCI .Experiment results indicate that the developed approach is efficient and can be used to natural non-wearable HCI .