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

Image Recognition and Understanding based on the Geometric Information Criterion

基于几何信息准则的图像识别与理解

基本信息

批准号：
09680352
负责人：
KANATANI Kenichi
金额：
$ 1.98万
依托单位：
GUNMA UNIVERSITY
依托单位国家：
日本
项目类别：
Grant-in-Aid for Scientific Research (C)
财政年份：
1997
资助国家：
日本
起止时间：
1997 至 1998
项目状态：
已结题

来源：
https://kaken.nii.ac.jp/grant/KAKENHI-PROJECT-09680352/
关键词：
image processing geometric AIC information criterion optical flow structure from motion 3-D reconstruction CAD mobile robot 画像理解コンピュータビジョンロボティクス

项目摘要

1. Self-evaluation from motion imagesWe have experimentally confirmed that the reliability of 3-D reconstruction from images taken by a moving camera can be evaluated from the original images by using the geometric AIC.2. Detection of degeneracyWe have experimentally confirmed that in binocular stereo vision and 3-D analysis from two images, the geometric AIC can well distinguish whether the object is a planar surface, the object is located infinitely far away, or the camera is simply rotating without translation.3. CAD interfaceWe have experimentally confirmed that regular figures manually input by humans can auto- matically be corrected into intended shapes without introducing empirical thresholds by the use of the geometric AIC.We have also demonstrated that curves and surfaces described by equations can be displayed efficiently by using interval analysis.4. Moving object detection from motion imagesWe have experimentally confirmed that we can detect objects moving in the scene by applying the geometric AIC to a image sequence of the scene taken by a moving camera.5. Rotation estimationWe have developed a technique for optimally computing the 3-D rotation of objects from their noisy stereo images and experimentally confirmed its effectiveness. 6. Robot self-localization We have developed a technique for optimally computing the current position of a mobile robot by matching a noisy image it takes with the 3-D model of the environment and experimentally confirmed its effectiveness.

1.从运动图像的自我评价我们已经通过实验证实，从运动相机拍摄的图像的3-D重建的可靠性可以通过使用几何AIC从原始图像进行评价。退化检测实验证实，在双目立体视觉和两幅图像的三维分析中，几何AIC可以很好地区分物体是否是平面，物体是否位于无限远的地方，或者相机是否只是简单地旋转而没有遮挡. CAD接口我们已经通过实验证实，使用几何AIC可以在不引入经验阈值的情况下，自动地将人工输入的规则图形校正为预期的形状.我们还证明了使用区间分析可以有效地显示由方程描述的曲线和曲面.从运动图像中检测运动目标我们已经通过实验证实，我们可以通过将几何AIC应用于由移动相机拍摄的场景的图像序列来检测场景中的运动目标.旋转估计我们已经开发了一种技术，用于最佳计算的3-D旋转的对象从他们的噪声立体图像，并通过实验证实了其有效性。6.机器人自我定位我们已经开发出一种技术，用于最佳地计算当前位置的移动的机器人匹配的噪声图像，它需要与3-D模型的环境和实验证实了其有效性。