公开(公告)号：KR100808543B1

公开(公告)日：2008-02-29

申请号：KR1020060089785

申请日：2006-09-15

Applicant: 한국과학기술연구원

Inventor： 박성기 ,  전세웅

IPC: G06T7/20 ,  G06T7/00 ,  G06T7/40 ,  G06T1/20

CPC classification number: G06T7/20 ,  G06K9/00335 ,  G06T5/40 ,  G06T7/44 ,  G06T7/85 ,  G06T17/005

Abstract: An apparatus and a method for acquiring human motion information by using a stereo vision sensor are provided to acquire motion information of a human by using a stereo camera, thereby analyzing interaction between human and robot and motions such as gestures of human under ubiquitous environment. A stereo camera(100) collects and provides a stereo image and a depth map for each pixel of an image. A background separation processing module(200) separates a human, corresponding to a foreground, from a background based on the depth map. A mechanical model module(300) for a human provides a three dimensional plane association model for human and two dimensional complex Gaussian model. A self-block separation module(400) divides the stereo image and disparity information for human, separated in the background separation processing module, at predetermined intervals according to a model provided from the mechanical model module of human, and separates self-blocks by generating a two dimensional image through histogram analysis for the disparity information at the interval. A motion analyzing module(500) acquired three dimensional motion of the human by analyzing human motions by comparing the two dimensional image with the two dimensional complex Gaussian model and then by performing fitting with the three dimensional plane association model by adding distance information.

Abstract translation: 提供了一种通过使用立体视觉传感器来获取人体运动信息的装置和方法，以通过使用立体摄像机来获取人的运动信息，由此分析人与机器人之间的相互作用以及运动，例如在普遍存在的环境下人的手势。 立体相机（100）收集并提供图像的每个像素的立体图像和深度图。 背景分离处理模块（200）基于深度图将与前景相对应的人与背景分离。 人类的机械模型模块（300）为人类和二维复数高斯模型提供了一个三维平面关联模型。 自组块分离模块（400）根据从人的机械模型模块提供的模型以预定的间隔划分在背景分离处理模块中分离的用于人的立体图像和视差信息，并且通过产生 通过对间隔信息的视差信息进行直方图分析的二维图像。 运动分析模块（500）通过将二维图像与二维复合高斯模型进行比较，分析人体运动，然后通过添加距离信息与三维平面关联模型进行拟合，从而获得人的三维运动。