CAJ | 학술논문

文章提出一种将运动和肤色特征相融合的多线索识别方法，利用差分运动检测降低复杂背景干扰，利用肤色在H S V空间中的聚类特性，采用多高斯模型，对每一个输入视频，将前50帧作为学习样本，在线学习出该视频的肤色分类器，解决了泛化性能低的问题；利用手势固有形状特征进行静态手势分割，采用基于仿射变换的块匹配法对动态手势进行跟踪。结果表明该算法能够较好地适应不同光照及复杂背景，同时满足实时要求。
문장제출일충장운동화부색특정상융합적다선색식별방법，이용차분운동검측강저복잡배경간우，이용부색재H S V공간중적취류특성，채용다고사모형，대매일개수입시빈，장전50정작위학습양본，재선학습출해시빈적부색분류기，해결료범화성능저적문제；이용수세고유형상특정진행정태수세분할，채용기우방사변환적괴필배법대동태수세진행근종。결과표명해산법능구교호지괄응불동광조급복잡배경，동시만족실시요구。
An online dynamic hand gesture recognition method with multiple cues based on the motion and the features of skin color is proposed .The disturbance caused by complex background is reduced by motion detection .According to the cluster characteristic of skin color in HSV space ,and by using the multi-Gaussian model ,the first 50 frames of each input video are used as the learning sample to construct the online skin classifier ,w hich can solve the low generalization performance problem .T he static hand recognition is completed with geometric feature .An affine model is adopted for the motion displacement estimation in hand tracking .The experimental results show that the proposed method is robust and real-time ,and is able to adapt to different illumination and complex background .