| Project/Area Number |
23KF0094
|
|---|
| Research Category |
Grant-in-Aid for JSPS Fellows
|
| Allocation Type | Multi-year Fund |
|---|
| Section | 外国 |
|---|
| Review Section |
Basic Section 61050:Intelligent robotics-related
|
|---|
| Research Institution | National Institute of Advanced Industrial Science and Technology |
|---|
Principal Investigator |
佐川 立昌 国立研究開発法人産業技術総合研究所, 情報・人間工学領域, 研究チーム長 (30362627)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
CAILLOT ANTOINE 国立研究開発法人産業技術総合研究所, 情報・人間工学領域, 外国人特別研究員
|
|---|
| Project Period (FY) |
2023-04-25 – 2025-03-31
|
| Project Status |
Granted (Fiscal Year 2023)
|
| Budget Amount *help |
¥2,000,000 (Direct Cost: ¥2,000,000)
Fiscal Year 2024: ¥1,000,000 (Direct Cost: ¥1,000,000)
Fiscal Year 2023: ¥1,000,000 (Direct Cost: ¥1,000,000)
|
|---|
| Keywords | Computer Vision / SLAM / High Dynamic Range |
|---|
| Outline of Research at the Start |
This research aims to improve the robustness of Simultaneous Localization and Mapping (SLAM) algorithms in severe lighting environments by integrating emergent sensors. The expected outcome is an enhanced robustness to various challenging environments.
|
| Outline of Annual Research Achievements |
This fiscal year was an opportunity to explore the impact of the challenging lighting conditions in visual Simultaneous Localization And Mapping (SLAM), especially when using a camera equipped with a fisheye lens, on mobile robots. Numerous situations have been found where the SLAM fails because of wide illumination differences in the images acquired. Therefore, a preprocessing layer creating a composite image from two exposures has been developed to feed a SLAM algorithm. Unlike other state-of-the-art methods, the fact that our system is aimed to be used on mobile robots prohibits sequential capture of multiple images differently exposed as it would result in artifacts in the composite image caused by the movement of the robot. This preprocessing layer enabled the usage of visual SLAM in places where the lighting condition prohibits the usage of classical image capture methods.
|
| Current Status of Research Progress |
Current Status of Research Progress
2: Research has progressed on the whole more than it was originally planned.
Reason
The research on this project has taken longer time than expected as the first tried approach was the deep modification of a state-of-the-art visual SLAM algorithm as well as the usage of the previously recorded datasets which did not feature strong enough contrast in the images. After determining that the taken approach would not lead to results within the time frame of the postdoctoral fellowship, we switched to another approach which highlighted the fact that the dataset we were using was not sufficient. The recording of new datasets and the usage of the second approach led to qualitative results: we succeeded in running visual SLAM in a challenging environment where it was impossible before. It now remains the obtention of the quantitative results. The research on hyperspectral visual SLAM had to be put on hold as the budget required for a camera exceeded the whole budget of this postdoctoral fellowship.
|
| Strategy for Future Research Activity |
In the short term, quantitative data will obtained in a wider range of situations, enabling the comparison of multiple methods to build HDR images and configuration of the visual SLAM. Then, the research will be extended with the usage of a multi-camera system enabling a wider contrast sensitivity in the image sensing, attaining the 120dB of dynamics required for everyday scenes (our current configuration allows up to 90dB). Finally, we expect to get hyperspectral datasets to run visual SLAM and improve feature extraction and quality.
|
