SR-LIVO

Re-segments LiDAR sweeps to end at image timestamps so that LIO alone estimates camera poses, reducing the vision module to camera-parameter refinement and point coloring.

技術屬性

欄位內容為文獻擷取紀錄的原文用語（英文），以原文為據；「未查證」表示本研究尚未讀到該資訊，不代表該方法不具備此能力。

使用設備

原文使用的感測器、運算硬體與載具（equipment）。型號保留原文寫法，連結到設備頁中同一型號的歸併名稱；角色依原文用途分為方法輸入、資料集感測器、執行運算平台、參考或真值量測（reference or ground truth）與比較對象設備。

作者報告的優勢與限制

優勢

• Lowest or tied-lowest RMSE ATE on 8 of 9 NTU-VIRAL sequences against R3Live and Fast-LIVO; Fast-LIVO is better on eee_02 (0.18 vs 0.23 m) (Table I)
• Higher PSNR and SSIM of the colorized map than R3Live on all 15 test sequences, e.g., eee_01 PSNR 18.92 vs 12.11 and SSIM 0.82 vs 0.37 (Table II)
• About 1.6 times faster than R3Live with comparable colored reconstruction on r3live_01 and a better grayscale map on eee_01 (Sec. VI-D to VI-F, Table VI, Figs. 4 to 6)
• Rendering with LIO poses at image times gives higher PSNR and SSIM than rendering with the authors' LiDAR-assisted VIO module on all sequences (Table IV)

限制

• Image rate must be down-sampled to at most twice the LiDAR rate, otherwise reconstructed sweeps become too sparse for LIO (Sec. V-A)
• No loop closure (Sec. VII)
• Premise holds only without LiDAR degeneration: R3Live sequences with LiDAR failure modes were deliberately excluded, and long sequences whose colored map exceeded 32 GB RAM were also excluded (Sec. VI)
• Trajectory accuracy is evaluated only on NTU-VIRAL because the R3Live sequences have no position ground truth (Sec. VI-A)
• The LIO-over-VIO claim is not uniform: Fast-LIVO's VIO output is more accurate than its LIO output on sbs_01 (0.42 vs 0.73 m) (Table III)

營建工程相關證據

原文未報告

原文驗證環境：公開基準

來源

• Yuan et al., 2024

  Zikang Yuan, Jie Deng, Ruiye Ming, Fengtian Lang, Xin Yang（2024）SR-LIVO: LiDAR-Inertial-Visual Odometry and Mapping With Sweep ReconstructionIEEE Robotics and Automation Letters, 9(6): 5110-5117

  DOI 10.1109/lra.2024.3389415arXiv 2312.16800程式碼

  同儕審查已出版已讀全文近十年

相關版本

• 預印本：SR-LIVO arXiv (2023-12-28) https://arxiv.org/abs/2312.16800
• 程式碼釋出：ZikangYuan/sr_livo https://github.com/ZikangYuan/sr_livo

程式碼：https://github.com/ZikangYuan/sr_livo（授權：GPL-2.0 (LICENSE file checked)）。有公開程式碼不等於已被重現，也不代表目前版本與論文版本相同。

回到方法圖鑑