Current Result Document :
| 한글제목(Korean Title) |
Residual transformation 추정 기반 Visual Odometry |
| 영문제목(English Title) |
Visual Odometry via Residual transformation regression |
| 저자(Author) |
김동욱
서승우
Dongwook Kim
Seung-Woo Seo
|
| 원문수록처(Citation) |
VOL 45 NO. 01 PP. 2210 ~ 2213 (2022. 06) |
한글내용
(Korean Abstract) |
|
영문내용
(English Abstract) |
Recent studies in deep learning show that deep neural networks can obtain depth and optical flow accurately. This enables deep learning based visual odometry can outperform traditional hand-crafted feature mathing based odometry algorithm. DF-VO achieved high accuracy on urban scenes such as KITTI odometry dataset, but its performance becomes lower in wild environment. Our paper proposed residual DF-VO, which uses rough prediction of transformation between two frames and regresses residual transformation matrix to compensate error. Experiments in Finnforest dataset show that out proposed algorithm can achieve higher accuracy that original DF-VO in wild environment. |
| 키워드(Keyword) |
|
| 파일첨부 |
PDF 다운로드
|
