Disparity estimation based on feature map correlation using contrastive learning
特徴マップの相関に基づいた対照学習による視差算出の高精度化
- Delivery
- Available on the other site
- Click here to order.
- Publication code
- 20244534
- Paper/Info type
- JSAE Transaction
Vol.55 No.5
- Pages
- 1021-1026(Total 6 p)
- Date of publication
- Sep 2024
- Publisher
- JSAE
- Language
- Japanese
Detailed Information
| Category(J) | 論文 Translation |
|---|---|
| Category(E) | Paper |
| Author(J) | 1) 二宮 洸, 2) 遠藤 健, 3) 城戸 英彰, 4) 入江 耕太 |
| Author(E) | 1) Takeru Ninomiya, 2) Takeshi Endo, 3) Hideaki Kido, 4) Kota Irie |
| Affiliation(J) | 1) 日立製作所, 2) 日立製作所, 3) 日立製作所, 4) 日立Astemo |
| Abstract(J) | 深層学習に基づく視差算出により高精度に奥行き情報を計算できるが，基線長が長いカメラペアでは左右画像の写り方の違いが大きくなり，視差の精度が低下する．そこで，特徴抽出において左右画像の見え方の違いを埋めるために対照学習を用いた学習方法を提案する．実験から，近傍20m以内における精度向上を確認した． Translation |
| Abstract(E) | Depth can be estimated with high accuracy by using deep learning. However, for camera pairs with long baseline, the accuracy of disparity is reduced because of the visual difference between left and right images. In this paper, we propose disparity estimation method based on feature map correlation using contrastive learning. By taking into account visual difference between the left and right images, we improve the accuracy of disparity estimation. Experimental results show that the proposed method improves accuracy within 20 m. |
