SLAM 點雲圖鑑Point clouds for construction
M-01 · 2025

FAST-LIVO2

FAST-LIVO2 以序列式更新的 ESIKF 先融合光達、再融合影像，解決兩種量測維度不匹配的問題；光達與視覺模組共用一個自適應體素（voxel）地圖，光達點同時作為視覺地圖點並附掛影像區塊。影像對齊利用光達平面先驗、動態更新參考區塊、按需射線投射（raycasting）處理近距盲區，並即時估計曝光時間。作者在 NTU-VIRAL 與 Hilti（含工地序列）等 25 個公開序列評估軌跡精度，並展示高精度彩色點雲、網格、貼圖與 3DGS 應用。

里程計＋局部建圖近十年已讀全文同儕審查已出版LiDAR、視覺與慣性融合

閱讀版本
arXiv 2408.14035v2 HTML (2024-08-28), all main sections I to XI; IEEE T-RO version of record (Xplore document 10757429) HTML text spot-checked for Sec. VIII-A, IX-A, IX-B, IX-E, X-A2, X-B, X-C and XI with matching numbers; T-RO table images returned HTTP 403, so tables are from arXiv v2; Supplementary Material not read
全文取得途徑
arXiv 預印本
本頁內容

A direct LIVO with a sequentially updated ESIKF over a single adaptive voxel map, using LiDAR plane priors, reference-patch updates, raycasting and exposure estimation to obtain pixel-level alignment and dense colored point maps.

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。

感測輸入3D LiDAR (Livox Avia, Ouster OS1-16, Hesai PandarXT-32, Robosense BPearl across datasets)、IMU、camera (pinhole or fisheye)
原文測試平台handheld、UAV、ground robot (Hilti robot-mounted sequences; type 未查證)
狀態估計error-state iterated Kalman filter with sequential update (LiDAR update first, then image update)
資料關聯raw LiDAR points, frame-to-map point-to-plane with per-point noise including beam divergence; sparse-direct patch photometric alignment using LiDAR plane priors, dynamic reference-patch update, on-demand voxel raycasting, outlier rejection
時間表示discrete poses; LiDAR points recombined into scans at camera sampling times; online exposure-time estimation
去畸變scan recombination with forward and backward IMU propagation (Fig. 3)
迴圈閉合none; authors note possible long-distance drift and list loop closure as future work (Sec. XI)
全域最佳化none
地圖表示Adaptive voxel map adapted from VoxelMap: hash table of 0.5 m root voxels, each an octree (max 3 layers) of plane leaf voxels with plane center, normal and covariance; mature planes stop accepting points; selected points carry 3-level patch pyramids (visual map points); local map of side L slid as a ring buffer when the detection sphere touches the boundary
先驗資訊hardware time synchronization and pre-calibrated extrinsics required (Sec. I)
可輸出幾何dense colored point map in real time; downstream TSDF mesh (VDBFusion), OpenMVS texture mapping and 3DGS initialization demonstrated (Sec. X-C)
計算需求Desktop Intel i7-10700K with 32 GB RAM: average 30.03 ms per LiDAR and image frame (17.13 ms LiDAR, 12.90 ms image) over all benchmark and private sequences; ARM RB5 (Qualcomm Kryo585, 8 GB): 78.44 ms average; onboard NUC i7-1360P during UAV flights about 53.47 ms while planning and MPC also run; MARS-LVIG airborne sequences about 25.2 ms and 21.8 ms

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。

類別型號（原文寫法）角色資料集原文規格出處
LiDAR16-channel OS1 gen1資料集感測器NTU-VIRAL10 Hz; built-in IMU at 100 Hz(Zheng et al., 2025, Sec. VIII-A)
LiDARHesai PandarXT-32資料集感測器Hilti'22 and Hilti'23 (handheld)10 Hz (handheld sequences)(Zheng et al., 2025, Sec. VIII-A)
LiDARRobosense BPearl歸入：Robosense Bpearl資料集感測器Hilti (robot-mounted)10 Hz (robot-mounted sequences)(Zheng et al., 2025, Sec. VIII-A)
LiDARLivox Avia資料集感測器MARS-LVIGwith built-in BMI088 IMU; triggered at 10 Hz(Zheng et al., 2025, Sec. VIII-A)
LiDARLivox Avia方法輸入FAST-LIVO2 private datasetFoV 70.4 x 77.2 deg(Zheng et al., 2025, Sec. VIII-B1; Fig. 9)
慣性量測單元（IMU）Bosch BMI085資料集感測器Hilti'22 and Hilti'23 (handheld)external IMU at 400 Hz(Zheng et al., 2025, Sec. VIII-A)
慣性量測單元（IMU）Xsens MTi-670資料集感測器Hilti (robot-mounted)200 Hz(Zheng et al., 2025, Sec. VIII-A)
相機two synchronized pinhole cameras (left used)資料集感測器NTU-VIRALtriggered at 10 Hz; 752x480 grayscale(Zheng et al., 2025, Sec. VIII-A)
相機five wide-angle cameras (front-facing used)資料集感測器Hilti'22 and Hilti'23 (handheld)40 Hz downsampled to 10 Hz; 752x480 grayscale(Zheng et al., 2025, Sec. VIII-A)
相機eight omnidirectional cameras (front-facing used)資料集感測器Hilti (robot-mounted)10 Hz(Zheng et al., 2025, Sec. VIII-A)
相機high-resolution global-shutter camera資料集感測器MARS-LVIG2448x2048 RGB, triggered at 10 Hz(Zheng et al., 2025, Sec. VIII-A)
相機MV-CA013-21UC方法輸入FAST-LIVO2 private datasetindustrial camera, FoV 70.6 x 68.5 deg; fixed exposure with auto gain in most sequences(Zheng et al., 2025, Sec. VIII-B1; Fig. 9)
全測站Total Station參考或真值量測Hilti'22 and Hilti'23millimeter-accurate ground truth; not public, scored online(Zheng et al., 2025, Sec. VIII-A)
載具平台aerial platform資料集感測器NTU-VIRALUAV campus flights(Zheng et al., 2025, Sec. VIII-A)
載具平台DJI M300 RTK quadrotor資料集感測器MARS-LVIGhigh-altitude aerial data collection(Zheng et al., 2025, Sec. VIII-A)
載具平台handheld platform方法輸入FAST-LIVO2 private datasethandheld data-collection device (Fig. 9a)(Zheng et al., 2025, Fig. 9)
運算硬體DJI manifold-2c歸入：DJI Manifold 2C資料集感測器FAST-LIVO2 private datasetonboard computer; Intel i7-8550u CPU, 8 GB RAM(Zheng et al., 2025, Sec. VIII-B1)
運算硬體desktop PC with Intel i7-10700K CPU執行運算平台未標示32 GB RAM(Zheng et al., 2025, Sec. IX-A)
運算硬體RB5 (ARM) with Qualcomm Kryo585 CPU執行運算平台未標示8 GB RAM(Zheng et al., 2025, Sec. IX-A)
運算硬體NUC with Intel i7-1360P CPU執行運算平台未標示32 GB RAM; UAV onboard computer(Zheng et al., 2025, Sec. X-A1)
其他motion capture system (MoCap)參考或真值量測Hilti'22 and Hilti'23millimeter-accurate ground truth; not public, scored online(Zheng et al., 2025, Sec. VIII-A)
其他STM32 synchronized timers資料集感測器FAST-LIVO2 private dataset10 Hz hardware trigger for all sensors(Zheng et al., 2025, Sec. VIII-B1; Fig. 9)

作者報告的優勢與限制

優勢

  • Average RMSE of about 0.044-0.045 m over 25 NTU-VIRAL and Hilti 2022/2023 sequences (text 0.044 m; default Table II row 0.045 m), versus 0.137 m for FAST-LIVO in the same table (Sec. IX-B, Table II)
  • End-to-end error below 0.01 m on several private degenerate sequences including a dim mining tunnel (Sec. IX-C)
  • Colored point maps suitable as inputs for mesh, texture and 3DGS pipelines (Sec. X-C)

限制

  • Odometry only
  • may drift over long distances without loop closure (Sec. XI) | Pixel-level accuracy requires hardware synchronization and rigorous extrinsic pre-calibration (Sec. I) | Slightly worse than FAST-LIO2 on dark, blurred sequences where images add no information (Sec. IX-B) | Normal refinement helps only in simple structured scenes with good images and degrades dim, blurry NTU-VIRAL sequences, so it is off by default (Sec. IX-B) | Denser colored points increase 3DGS training time from 10 min 59 s to 15 min 30 s compared with COLMAP input (Sec. X-C)

營建工程相關證據

作者以 Hilti 2022 與 2023 公開資料集評估軌跡精度，論文描述其場景含營建工地、辦公室、地下室與樓梯，參考軌跡由動作捕捉或全測站取得且不公開，需經 Hilti 官方網站評分。Hilti 2022 三個工地序列 Construction Ground、Construction Multilevel、Construction Stairs 的 RMSE 分別為 0.010、0.020、0.016 m。私人資料含礦坑隧道序列，僅報告回到起點誤差小於 0.01 m。證據屬軌跡層級，未報告工地點雲的幾何精度。

原文驗證環境：公開基準、施工中工地、地下或隧道、獨立參考量測

報告的性能數據

以下是原文作者報告的性能數值（author-reported results），不是本研究重新量測的結果。每張圖只並列同一個比較組（comparison group，同一張表、同一組實驗設定）內的方法；不同比較組之間的數值不可直接比較，也不構成排名。

本方法共出現在 12 個比較組，合計 103 筆紀錄。以下列出本方法紀錄最多的 4 組，其餘 8 組列在最後，並連到性能比較頁。

Tang et al., 2026 · Table II 本方法 21 筆

指標absolute translation error (RMSE, meters)

表格設定（擷取紀錄原文）：Absolute translation error (RMSE, m), all systems in real-time mode on the desktop PC; 'x' = system totally failed. Ablation columns (Ours VIO, LIO, w/o F2M, Marg. F2M, w/o calib.) omitted here. MARS-LVIG ground truth re-derived by the authors with post-processed RTK/INS; i2Nav-Robot ground-truth source not described in this paper. Average rows printed by the authors. (Tang et al., 2026, Table II)

absolute translation error (RMSE, meters)，i2Nav-Robot · building00

只並列這張表在相同設定下報告的方法；以「本方法：」開頭者為本頁方法。失敗、未執行與未報告以標記呈現，不是 0。

  • 失敗

這些是 Tang et al., 2026 在此表設定下報告的數值（author-reported results），只能在同一個比較組內對照，不代表方法在其他資料或設定下的表現。

統計量：均方根誤差（RMSE）；對齊方式：原文未報告；單位：m；場景：low-speed wheeled robot; building, parking, playground and street sequences (indoor-outdoor)

資料來源作者報告值(Tang et al., 2026, Table II)

數值與出處
方法（原文寫法）報告值出處
FF-LINS2.32 m(Tang et al., 2026, Table II)
FAST-LIO20.68 m(Tang et al., 2026, Table II)
LE-VINS1.17 m(Tang et al., 2026, Table II)
R3LIVE無數值失敗註記（擷取紀錄）：failed(Tang et al., 2026, Table II)
FAST-LIVO2本方法1.1 m(Tang et al., 2026, Table II)
Ours (PA-LVIO)原文提出0.34 m(Tang et al., 2026, Table II)

Zheng et al., 2025 · Table II 本方法 17 筆

資料集與序列NTU-VIRAL, Hilti'22, Hilti'23 · Average (25 sequences)

表格設定（擷取紀錄原文）：Absolute translational error RMSE on Hilti'22 and Hilti'23 (handheld: PandarXT-32, BMI085; robot: BPearl, MTi-670; front camera), scored through the official Hilti website because ground truth (MoCap or total station) is not public; loop closure of LVI-SAM removed; ablation columns kept only in the Average row; NTU-VIRAL rows omitted for the row cap; Average over all 25 sequences (NTU-VIRAL, Hilti'22, Hilti'23; how failed runs enter the average is not stated) (Zheng et al., 2025, Table II)

absolute translational errors (RMSE), Average row，NTU-VIRAL, Hilti'22, Hilti'23 · Average (25 sequences)

只並列這張表在相同設定下報告的方法；以「本方法：」開頭者為本頁方法。失敗、未執行與未報告以標記呈現，不是 0。

這些是 Zheng et al., 2025 在此表設定下報告的數值（author-reported results），只能在同一個比較組內對照，不代表方法在其他資料或設定下的表現。

統計量：均方根誤差（RMSE）；對齊方式：原文未報告；單位：m；場景：mixed aerial, construction, indoor and outdoor

資料來源作者報告值(Zheng et al., 2025, Table II)

數值與出處
方法（原文寫法）報告值出處
SDV-LOAM7.416 m(Zheng et al., 2025, Table II (Average row); text Sec. IX-B states 0.044 m for Ours)
Our LIO0.097 m(Zheng et al., 2025, Table II (Average row); text Sec. IX-B states 0.044 m for Ours)
FAST-LIO20.151 m(Zheng et al., 2025, Table II (Average row); text Sec. IX-B states 0.044 m for Ours)
R3LIVE0.278 m(Zheng et al., 2025, Table II (Average row); text Sec. IX-B states 0.044 m for Ours)
LVI-SAM1.928 m(Zheng et al., 2025, Table II (Average row); text Sec. IX-B states 0.044 m for Ours)
FAST-LIVO0.137 m(Zheng et al., 2025, Table II (Average row); text Sec. IX-B states 0.044 m for Ours)
Ours本方法原文提出0.045 m(Zheng et al., 2025, Table II (Average row); text Sec. IX-B states 0.044 m for Ours)
Ours (w/o expo)0.051 m(Zheng et al., 2025, Table II (Average row); text Sec. IX-B states 0.044 m for Ours)
Ours (w normal)0.044 m(Zheng et al., 2025, Table II (Average row); text Sec. IX-B states 0.044 m for Ours)
Ours (w/o update)0.089 m(Zheng et al., 2025, Table II (Average row); text Sec. IX-B states 0.044 m for Ours)

Zhou et al., 2025 · Table I 本方法 17 筆

表格設定（擷取紀錄原文）：ATE RMSE on 16 Hilti'22 and Hilti'23 sequences computed with the official Hilti evaluation tools; parameters of all methods tuned by the authors; x = system totally failed (Zhou et al., 2025, Table I)

ATE (RMSE)，Hilti'22 · Construction Ground

只並列這張表在相同設定下報告的方法；以「本方法：」開頭者為本頁方法。失敗、未執行與未報告以標記呈現，不是 0。

  • 失敗

這些是 Zhou et al., 2025 在此表設定下報告的數值（author-reported results），只能在同一個比較組內對照，不代表方法在其他資料或設定下的表現。

統計量：均方根誤差（RMSE）；對齊方式：原文未報告；單位：m；場景：construction site

資料來源作者報告值(Zhou et al., 2025, Table I)

數值與出處
方法（原文寫法）報告值出處
Ours原文提出0.01 m(Zhou et al., 2025, Table I)
FAST-LIVO2本方法0.01 m(Zhou et al., 2025, Table I)
FAST-LIO20.013 m(Zhou et al., 2025, Table I)
FAST-LIVO0.022 m(Zhou et al., 2025, Table I)
R3LIVE0.021 m(Zhou et al., 2025, Table I)
SDV-LOAM25.121 m(Zhou et al., 2025, Table I)
LVI-SAM無數值失敗註記（擷取紀錄）：failed (x)(Zhou et al., 2025, Table I)

Tao et al., 2025 · Table 3 本方法 14 筆

指標RMS of ATE

表格設定（擷取紀錄原文）：ATE RMS (m) against LiDAR-to-TLS ground truth after SE(3) Umeyama alignment; online: VILENS-SLAM, Fast-LIO-SLAM, SC-LIO-SAM, ImMesh, Fast-LIVO2; offline: HBA (input VILENS-SLAM), COLMAP (images only). VILENS-SLAM = VILENS with pose-graph optimisation; Fast-LIO-SLAM and SC-LIO-SAM add Scan Context loop closures to Fast-LIO2 and LIO-SAM. 'x' in the table = failed or incomplete. Authors note methods could improve with further tuning. (Tao et al., 2025, Table 3)

RMS of ATE，Oxford Spires · Keble College 02 (290 m)

只並列這張表在相同設定下報告的方法；以「本方法：」開頭者為本頁方法。失敗、未執行與未報告以標記呈現，不是 0。

這些是 Tao et al., 2025 在此表設定下報告的數值（author-reported results），只能在同一個比較組內對照，不代表方法在其他資料或設定下的表現。

統計量：均方根誤差（RMSE）；對齊方式：SE(3) 剛體對齊；單位：m；場景：historic site, outdoor and indoor parts (Keble College, Oxford)

資料來源作者報告值(Tao et al., 2025, Table 3)

數值與出處
方法（原文寫法）報告值出處
VILENS-SLAM0.06 m(Tao et al., 2025, Table 3)
Fast-LIO-SLAM0.25 m(Tao et al., 2025, Table 3)
SC-LIO-SAM1.26 m(Tao et al., 2025, Table 3)
ImMesh0.08 m(Tao et al., 2025, Table 3)
Fast-LIVO2本方法0.95 m(Tao et al., 2025, Table 3)
HBA0.11 m(Tao et al., 2025, Table 3)
COLMAP0.05 m(Tao et al., 2025, Table 3)

其他比較組

列出其餘 8 個比較組

來源

  • Zheng et al., 2025

    Chunran Zheng, Wei Xu, Zuhao Zou, Tong Hua, Chongjian Yuan, Dongjiao He, Bingyang Zhou, et al.（2025）FAST-LIVO2: Fast, Direct LiDAR–Inertial–Visual OdometryIEEE Transactions on Robotics, 41: 326-346

    DOI 10.1109/tro.2024.3502198arXiv 2408.14035程式碼

    同儕審查已出版已讀全文近十年查證後修正

程式碼：https://github.com/hku-mars/FAST-LIVO2（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

SLAM 點雲圖鑑

本站為綜述論文《SLAM-Based Point Clouds for Construction: A Review》研究過程的互動說明文件。互動圖為教學模擬，不代表任何特定系統的實測表現；方法描述、數值與結論以所引原始文獻為準。

內容基準日 2026-09-25 · 研究與網站製作使用 AI 輔助工具，見 證據標記與方法

選擇開啟Esc關閉