CAJ | 학술논문

自然场景或电视剧中的人物，其姿势、背景、轮廓、大小总是不断发生变化，人体姿势估计是计算机视觉中具有挑战性的课题之一。针对静态图像中的人体姿态估计容易受背景的影响而又没有合适的背景提取方法，先大致确定人体范围，将调整的检测框的大小作为grabcut的输入，利用grabcut的分割结果确定出人所在的前景，进一步利用图像的边缘信息和区域信息推断出人体的姿势，最后通过实验证明了该方法的有效性。
자연장경혹전시극중적인물，기자세、배경、륜곽、대소총시불단발생변화，인체자세고계시계산궤시각중구유도전성적과제지일。침대정태도상중적인체자태고계용역수배경적영향이우몰유합괄적배경제취방법，선대치학정인체범위，장조정적검측광적대소작위grabcut적수입，이용grabcut적분할결과학정출인소재적전경，진일보이용도상적변연신식화구역신식추단출인체적자세，최후통과실험증명료해방법적유효성。
Human pose estimation is considered as one of the changeable problems in computer vision. In order to solve the problem that pose estimation is vulnerable to the background, we first detect the rough location of person in image, and then adjust the bounding box to make it cover all the people. The adjusted bounding box is seen as input of grabcut which is used to further segment the foreground and background, we combine the edge and region information to estimate human pose. Finally, the proposed method is proved by experiments.