-
公开(公告)号:KR1020050020217A
公开(公告)日:2005-03-04
申请号:KR1020030057979
申请日:2003-08-21
Applicant: 대한민국(전남대학교총장)
IPC: G06F9/46
Abstract: PURPOSE: A method for intelligently detecting/tracking a face in real-time through an active camera using a cascade classifier is provided to satisfy calculation efficiency and detection performance at the same time by using the cascade classifier, and accept diversity such as lighting change, complex background, overlap/tilt of the face, and face area change. CONSTITUTION: An integrated image is generated. A hierarchically formed cascade classifier is generated. The face is tracked by using face position information tracked in real-time. The lighting change is accepted by applying a Gaussian filter and a comparing extension method. The overlap between the complex background and the face region, and profile are accepted by using a feature mask to extract features of the face. The tile of the face is accepted by using the rotated feature mask. Diverse face area sizes are accepted by performing scanning after forming a pyramid having multiple resolutions.
Abstract translation: 目的:提供一种通过使用级联分类器通过主动摄像机实时智能检测/跟踪人脸的方法,以通过使用级联分类器同时满足计算效率和检测性能,并接受诸如照明变化等多样性, 复杂的背景,脸部的重叠/倾斜以及面部区域变化。 构成:生成集成图像。 生成层级形成的级联分类器。 通过使用实时跟踪的脸部位置信息跟踪脸部。 通过应用高斯滤波器和比较扩展方法可以接受照明变化。 通过使用特征掩码来提取面部特征,可以接受复杂背景与脸部区域和轮廓之间的重叠。 使用旋转的功能面罩可以接受脸部的瓷砖。 通过在形成具有多个分辨率的金字塔之后执行扫描可以接受不同的面部区域大小。