科研速览 · Science Skim继续刷下去 · Keep skimming →
◆ Cybernetics & Systems2025-12-26· Computer vision

VinsFusion-Line: Binocular Vision Inertial Navigation Real-Time SLAM System Based on Line Features

Guanghai Yang, Yulian Jiang, Shenquan Wang, Kun Chen

原始摘要(英文原文)· Original abstract
For the visual-inertial system, the key frame loss always emerges in the rotation with the weak texture and strong illumination, which seriously affects the camera positioning accuracy. To solve this problem, a tight-coupling-based visual fusion scheme leveraging line features is proposed in our work, called VinsFusion-Line. It first deeply integrates line features, inertial measurement unit(IMU) pre-integration, and stereo visual observations through a tightly-coupled framework. And the infinitely extended line features provide more geometric and structural information, compared with the base point and line feature tracing methods. Hence, the positioning accuracy is improved greatly. In order to further make the calculation of 3D spatial line features faster and more efficient, Plücker coordinates and straight lines are used to represent the features. In addition, the camera state is optimized by a multi-source cost function that embeds both IMU pre-integration residuals and line feature reprojection errors. This design enables tight-coupled fusion of IMU and binocular visual information, rather than relying solely on point features as in existing methods. Finally, experiments on the public data show that under the strong light and weak texture, the binocular VinsFusion-Line system with infinite extension line in our work has better positioning accuracy than the single-purpose Vins-Fusion system.
读原文 · Read the paper ↗

AI 追问PRO

登录后使用 AI 追问

讨论区

登录后参与讨论

相关论文 · Related

VinsFusion-Line: Binocular Vision Inertial Navigation Real-Time SLAM System Based on Line Features — 科研速览 Science Skim