| 제목 |
딥러닝 기반 Keypoint/Descriptor 추출 및 Optical Flow 매칭을 통한 Visual SLAM 성능 향상 연구 |
| 분야 |
전기ㆍ전자ㆍ통신 |
| 언어 |
Korean |
| 저자 |
이영현(국민대학교), 유진우(국민대학교) |
| Key Words |
SLAM(동시적 위치추정 및 지도 작성), Deep learning(딥러닝), Keypoint(특징점), Descriptor(기술자), Repeatability(반복성), Reliability(신뢰성), Optical flow(광류) |
| 초록 |
In this paper, we are proposing a methodology that can replace handcrafted feature points and descriptors by using deep learning, and enhance image matching by using optical flow. To obtain feature points with robust repeatability and reliability, we will utilize R2D2, a deep learning network based on L2-Net. The inferred feature points and descriptors are only passed to SLAM if they satisfy the minimum requirements for repeatability and reliability. Moreover, in image areas where there is optical flow, matching is performed by relying on optical flow. Experimental results show that our methodology achieves lower path errors in terms of RMSE for the most part of the experimental data than the existing ORB Extractor in ORB SLAM. |
| 원문(PDF) |
다운로드
Journal Site
|
