| Project/Area Number |
17500112
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
Perception information processing/Intelligent robotics
|
|---|
| Research Institution | OKYAMA UNIVERSITY |
|---|
Principal Investigator |
KANATANI Kenichi Okayama University Department of Computer Science, Professor, 自然科学研究科, 教授 (60125838)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
YASUAKI Sugaya Toyohashi University of Technology Department of Information and Computer Sciences, Lecture, 情報工学系, 講師 (00335580)
|
|---|
| Project Period (FY) |
2005 – 2006
|
| Project Status |
Completed (Fiscal Year 2006)
|
| Budget Amount *help |
¥3,500,000 (Direct Cost: ¥3,500,000)
Fiscal Year 2006: ¥1,500,000 (Direct Cost: ¥1,500,000)
Fiscal Year 2005: ¥2,000,000 (Direct Cost: ¥2,000,000)
|
|---|
| Keywords | 360° panorama / triangular mesh optimization / statistical optimization / hyperaccurate fitting / 3-D shape reconstruction / fundamental matrix / factorization / self-calibration / 多面体表示 / カメラモデル / 幾何学的当てはめ / 曲線当てはめ |
|---|
| Research Abstract |
1. High-density high-accuracy display of 3-D scenes
3-D polyhedral shapes reconstructed from images may be different from their true shapes. We established a new technique for detecting the discrepancy by analyzing the input images, by which more realistic display is made possible.
Displaying multiple images of the scene around the viewer may cause inconsistencies when the viewing direction is moved by 360° or more. We devised a technique for optimally pasting images subject to the condition that no inconsistency occurs, by which the scene is continuously displayed for arbitrary changes of the viewing direction.
2. Theory for high accuracy geometric fitting and its applications
We gave a theoretical foundation to the "renormalization" method, which this investigator proposed before with a worldwide impact, in relation to the theoretical accuracy bound. We also analyzed its accuracy, which was known only to a first approximation, up to second order terms strictly. This lead to a discovery o
… More
f a "hyperaccurate" method that outperforms all existing methods. We confirmed this by experiments.
3. Efficient and high-accuracy computation of the fundamental matrix from two images
We devised a new technique for computing the fundamental matrix from noisy point correspondence data, which is the first step of 3-D reconstruction from images. This method is more accurate and efficient than all existing methods.
4. High-accuracy and efficient technique for 3-D reconstruction from video streams
We extended the "factorization" method, which efficiently reconstructs the 3-D shape of the scene from feature point tracking over a video stream using affine camera modeling, to a general form which includes all existing models. We confirmed using simulated and real video images that an appropriate camera model is automatically selected by this method.
We also devised a high-accuracy "self-calibration" technique, to which we incorporated schemes for avoiding redundancies and predicting the values to be computed. We confirmed that the computation speed is increased by several thousand times. Less
|
