权益分类	功能权益	普通用户	{{item.name}}会员
{{category.name}}	{{benefitItem.name}}

Research on Active Stereo Vision System

主动立体视觉系统研究

基本信息

批准号：
04650378
负责人：
MIYAZAKI Fumio
金额：
$ 1.28万
依托单位：
Osaka University
依托单位国家：
日本
项目类别：
Grant-in-Aid for General Scientific Research (C)
财政年份：
1992
资助国家：
日本
起止时间：
1992 至 1993
项目状态：
已结题

来源：
https://kaken.nii.ac.jp/en/grant/KAKENHI-PROJECT-04650378/
关键词：
Binocular Stereo Active Motion Binocular Disparity Motion Parallax Occlusion Occuding Contour

项目摘要

This reseatch investigated the active stereo vision system to acquire the reliable 3D information of the environment efficiently. We debeloped the hardware and software of the Bision Robot which can control the movement of the stereo camera. We proposed an active method which can detect both raliable binocular disparities and occlusion efficiently, because the method solves the correspondence problem with out using the heuristic constraints. Furthermore, we extended the method so as to deal with the more complicated scene shich contains the curved objects. In general, the projected image of the bundary of the curved object is called Occluding Contour shich does not exist in physical. The conventional stereo method can not deal with the scene containing the curved objects, because the matching of these occluding contour produce the error in 3D information. We proposed a reliable method to hetect occluding contours and occlusion by active binocular stereo with an occluding contour model that describes the relation between the cotour of curved object and its projected image. Applying the image flow generated by moving one camera to the model, we can restrict possible matched points seen by the other camera. We detect occluding contours by fitting these matched points to the model. This method can find occluding contours and occlusion more reliably than conventional ones because stereo matching is performed by using the geometric constraint based on the occluding contour model instead of heuristic constraints.

为了有效地获取可靠的环境三维信息，本研究对主动立体视觉系统进行了研究。设计了一种能够控制立体摄像机运动的仿生机器人的硬件和软件。提出了一种主动的视差检测方法，该方法不需要使用启发式约束就能有效地检测出视差和遮挡。并将该方法推广到包含曲面物体的复杂场景。一般来说，曲面物体的轮廓线的投影像称为遮挡轮廓线，这种遮挡轮廓线在物理上是不存在的。传统的立体方法不能处理含有曲面物体的场景，因为这些遮挡轮廓的匹配会产生三维信息的误差。提出了一种可靠的方法来检测主动双目立体的遮挡轮廓和遮挡，该方法采用描述曲面物体轮廓与其投影图像之间关系的遮挡轮廓模型。将移动一台相机生成的图像流应用于模型，我们可以限制另一台相机看到的可能匹配点。我们通过将这些匹配点拟合到模型来检测遮挡轮廓。该方法利用基于遮挡轮廓模型的几何约束代替启发式约束进行立体匹配，比传统方法能更可靠地发现遮挡轮廓和遮挡。