SLAM 點雲圖鑑Point clouds for construction
M-01 · 2018

VINS-Mono

VINS-Mono 以單眼相機加低成本 IMU 估計具公制尺度的六自由度狀態，先以僅視覺 SfM 與視覺慣性對齊完成初始化（陀螺儀偏差、速度、重力方向與尺度），再以滑動視窗緊耦合融合 IMU 預積分（pre-integration）與特徵觀測。迴圈偵測後進行緊耦合重定位，並因 roll、pitch 可由 VIO 觀測，只對 x、y、z 與偏航角做 4-DOF 位姿圖最佳化。作者將稠密建圖列為未來工作。

具全域修正的完整 SLAM近十年已讀全文同儕審查已出版視覺、視覺慣性與 RGB-D 稠密 SLAM

閱讀版本
version of record, IEEE T-RO 34(4):1004-1020 (IEEE Xplore HTML full text, free access; Tables I and II read from the table images); arXiv 1708.03852v1 (2017-08-13, submitted version) also read and differs
全文取得途徑
國立臺灣大學機構訂閱
本頁內容

VINS-Mono provides robust initialisation and tightly-coupled sliding-window monocular VIO with IMU pre-integration, relocalization and 4-DOF pose-graph optimisation, yielding metric trajectories but only sparse 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monocular camera、IMU
原文測試平台UAV (EuRoC MAV、self-developed quadrotor)、handheld (camera plus DJI A3 IMU suite、VI-Sensor)、mobile phone (iPhone, VINS-Mobile) (Sec. IX)
狀態估計tightly-coupled sliding-window nonlinear optimisation with IMU pre-integration and marginalisation; 4-DOF pose-graph optimisation for global consistency (Sec. III, VI, VIII)
資料關聯KLT tracking keeping 100 to 300 uniformly spaced corners, RANSAC fundamental-matrix rejection; loop detection with DBoW2 on 500 extra BRIEF corners, then BRIEF matching with 2D-2D fundamental and 3D-2D PnP RANSAC checks before relocalization (Sec. IV-A, VII-A, VII-B)
時間表示discrete keyframe states with IMU pre-integration between frames; IMU-rate forward propagation (Sec. IV-B, VI-F)
去畸變不適用
迴圈閉合DBoW2 loop detection with temporal and geometric checks, tightly-coupled relocalization, and loop edges in the pose graph with Huber norm (Sec. VII, VIII-B)
全域最佳化4-DOF (x, y, z, yaw) pose-graph optimisation with Huber-weighted loop edges, run in a separate thread; merging of multiple sessions; five merged EuRoC MH sequences gave 0.21 m ATE RMSE over about 500 m (Sec. VIII-A to D, IX-A-2)
地圖表示sparse features in sliding window; keyframe pose graph
先驗資訊no prior map required; a previously saved pose graph (keyframe poses, loop links and BRIEF features) can be loaded and merged with the current map by 4-DOF pose graph optimization (Sec. VIII-D to F); camera-IMU extrinsics calibrated online (Sec. VI-A, X)
可輸出幾何metric 6-DOF trajectory and sparse features; dense mapping named as future work (Sec. X)
計算需求Intel i7-4790 3.60 GHz on the campus dataset: feature detection 15 ms and KLT 5 ms at 25 Hz, window optimization 50 ms at 10 Hz, loop detection 100 ms, pose graph optimization 130 ms (Table II); onboard Intel i7-5500U 3.00 GHz on the aerial robot with 100 Hz IMU-propagated output (Sec. IX-C-1); motion-only optimization about 5 ms versus more than 50 ms for full VIO on embedded computers (Sec. VI-E)

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
行動掃描設備iPhone (VINS-Mobile)方法輸入未標示30 Hz images at 640x480; arXiv v1 names an iPhone7 Plus(Qin et al., 2018, Sec. IX-C-2)
慣性量測單元（IMU）ADIS16448資料集感測器EuRoC200 Hz, synchronized(Qin et al., 2018, Sec. IX-A-1)
慣性量測單元（IMU）DJI A3 flight controller built-in IMU (ADXL278 and ADXRS290)方法輸入未標示100 Hz(Qin et al., 2018, Sec. IX-B-1, Fig. 18)
慣性量測單元（IMU）DJI A3 flight controller (ADXL278 and ADXRS290)方法輸入未標示100 Hz; also used for attitude stabilization control(Qin et al., 2018, Sec. IX-C-1, Fig. 21)
慣性量測單元（IMU）InvenSense MP67B方法輸入未標示built-in six-axis gyroscope and accelerometer of the iPhone, 100 Hz(Qin et al., 2018, Sec. IX-C-2)
相機MatrixVision mvBlueFOX-MLC200w方法輸入未標示forward-looking global shutter, 752x480, 20 Hz; hand-held indoor suite(Qin et al., 2018, Sec. IX-B-1, Fig. 18)
相機MatrixVision mvBlueFOX-MLC200w with 190-degree fisheye lens方法輸入未標示forward-looking global shutter, 752x480; MEI camera model(Qin et al., 2018, Sec. IX-C-1, Fig. 21)
雙目相機Aptina MT9V034資料集感測器EuRoCglobal shutter, WVGA monochrome, 20 FPS; only the left camera used(Qin et al., 2018, Sec. IX-A-1)
雙目相機VI-Sensor方法輸入HKUST campus datasethand-held; 25 Hz images and 200 Hz IMU; 5.62 km HKUST campus loop, 1 h 34 min(Qin et al., 2018, Sec. IX-B-2)
全測站Leica MS50參考或真值量測EuRoCground-truth states(Qin et al., 2018, Sec. IX-A-1)
載具平台self-developed aerial robot (quadrotor)方法輸入未標示tracks a figure-eight pattern of 1.0 m radius circles(Qin et al., 2018, Sec. IX-C-1, Fig. 21)
運算硬體Intel i7-4790執行運算平台未標示3.60 GHz; campus dataset timing (Table II)(Qin et al., 2018, Sec. IX-B-2)
運算硬體Intel i7-5500U執行運算平台未標示3.00 GHz onboard the aerial robot(Qin et al., 2018, Sec. IX-C-1)
其他VICON歸入：Vicon參考或真值量測EuRoCground-truth states(Qin et al., 2018, Sec. IX-A-1)
其他OptiTrack參考或真值量測未標示motion capture ground truth for the flight(Qin et al., 2018, Sec. IX-C-1)

作者報告的優勢與限制

優勢

  • keep
  • add: VINS_loop had the lowest or tied-lowest EuRoC ATE RMSE of the three compared configurations (OKVIS monocular, VINS, VINS_loop) in 9 of 11 sequences (Table I)
  • loop-closure-disabled flight drifted 0.29 % over 61.97 m (Sec. IX-C-1)

限制

  • Monocular VINS may reach weakly observable or degenerate conditions depending on motion and environment (Sec. X)
  • Dense map production needs further research (Sec. X)

營建工程相關證據

論文未報告營建工地測試；評估包含 EuRoC、室內與校園大範圍實驗。退化運動（例如等速直線）下的可觀測性問題與工地手持或載具掃描相關（推論）。

原文驗證環境：公開基準、受控實驗

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 27 個比較組，合計 231 筆紀錄。以下列出本方法紀錄最多的 4 組，其餘 23 組列在最後，並連到性能比較頁。

von Stumberg & Cremers, 2022 · Table II 本方法 29 筆

表格設定（擷取紀錄原文）：TUM-VI (handheld); RMSE ATE in m; other methods from the TUM-VI paper, DM-VIO median of 5 runs with SE(3) alignment; X = failure; sequence length in brackets (von Stumberg & Cremers, 2022, Table II)

RMSE ATE，TUM-VI · corridor1 (305 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 von Stumberg & Cremers, 2022 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：handheld, TUM-VI (large-scale indoor and outdoor scenes; the scene of each sequence is not described in this paper)

資料來源作者報告值(von Stumberg & Cremers, 2022, Table II)

數值與出處
方法（原文寫法）報告值出處
ROVIO (stereo)0.47 m(von Stumberg & Cremers, 2022, Table II)
VINS (VINS-Mono, mono)本方法0.63 m(von Stumberg & Cremers, 2022, Table II)
OKVIS (stereo)0.33 m(von Stumberg & Cremers, 2022, Table II)
BASALT (stereo)0.34 m(von Stumberg & Cremers, 2022, Table II)

Lin et al., 2021 · Table I 本方法 24 筆

表格設定（擷取紀錄原文）：Version of record Table I: relative rotation error (RRE, deg) and relative translation error (RTE, %) over all sub-sequences of each length, two fast-rotating handheld sequences (130 and 200 deg/s, mapping to (a) and (b) not stated) with D-GPS RTK ground truth; rows R2LIVE-LIO, R2LIVE-VIO and R2LIVE-LC (the latter undefined in the text) omitted here. Supersedes the median values in the arXiv v1 Fig. 11 caption. (Lin et al., 2021, Table I)

RRE (deg) over 50 m sub-sequences，R2LIVE Experiment-4 (authors' data, D-GPS RTK) · Experiment-4 (a), 50 m sub-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 Lin et al., 2021 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：原文未報告；對齊方式：原文未報告；單位：deg；場景：outdoor handheld, fast rotation

資料來源作者報告值(Lin et al., 2021, Table I)

數值與出處
方法（原文寫法）報告值出處
VINS-Mono本方法0.24 deg(Lin et al., 2021, VoR Table I)
Fast-Lio0.34 deg(Lin et al., 2021, VoR Table I)
Camvox0.67 deg(Lin et al., 2021, VoR Table I)
R2LIVE原文提出0.25 deg(Lin et al., 2021, VoR Table I)

Lin & Zhang, 2022 · Table III 本方法 24 筆

表格設定（擷取紀錄原文）：Relative rotation error (RRE, deg) and relative translation error (RTE, %) over all sub-sequences of 50 to 300 m in two seaport sequences (Belcher Bay Promenade) with D-GPS RTK ground truth; R3LIVE-HiRes uses 1280x1024 images and 0.01 m map point spacing, R3LIVE-RT 320x256 images and 0.10 m; LVI-SAM run with a modified LiDAR front-end for the Livox Avia; identical values in arXiv v1 and the version of record (Lin & Zhang, 2022, Table III)

RRE (deg) over 50 m sub-sequences，R3LIVE Experiment-3 (authors' data, D-GPS RTK) · Experiment-3 (a), 50 m sub-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 Lin & Zhang, 2022 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：原文未報告；對齊方式：原文未報告；單位：deg；場景：seaport promenade with pedestrians and open sea (handheld)

資料來源作者報告值(Lin & Zhang, 2022, Table III)

數值與出處
方法（原文寫法）報告值出處
R3LIVE-HiRes原文提出0.99 deg(Lin & Zhang, 2022, VoR Table III)
R3LIVE-RT原文提出1.48 deg(Lin & Zhang, 2022, VoR Table III)
LVI SAM2.11 deg(Lin & Zhang, 2022, VoR Table III)
R2LIVE1.21 deg(Lin & Zhang, 2022, VoR Table III)
FAST-LIO21.36 deg(Lin & Zhang, 2022, VoR Table III)
VINS-Mono本方法3.03 deg(Lin & Zhang, 2022, VoR Table III)

Rosinol et al., 2020 · Table II 本方法 22 筆

指標RMSE ATE [m]

表格設定（擷取紀錄原文）：EuRoC ATE RMSE grouped as fixed-lag smoothing, full smoothing and PGO with loop closure; comparator values taken from Delmerico and Scaramuzza [77] (Sim(3) alignment per text) and VINS-Mono [24]; comparators use a monocular camera while Kimera uses stereo; Kimera aligned with SE(3); loop threshold alpha = 0.001 (Rosinol et al., 2020, Table II)

RMSE ATE [m]，EuRoC MAV · MH_01

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 Rosinol et al., 2020 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：EuRoC MAV sequences (micro aerial vehicle dataset, ref. [19]); environments not described in this paper

資料來源作者報告值(Rosinol et al., 2020, Table II)

數值與出處
方法（原文寫法）報告值出處
OKVIS0.16 m(Rosinol et al., 2020, Table II)
MSCKF0.42 m(Rosinol et al., 2020, Table II)
ROVIO0.21 m(Rosinol et al., 2020, Table II)
VINS-Mono本方法0.15 m(Rosinol et al., 2020, Table II)
SVO-GTSAM (full smoothing)0.05 m(Rosinol et al., 2020, Table II)
VINS-LC (loop closure)本方法0.12 m(Rosinol et al., 2020, Table II)

其他比較組

列出其餘 23 個比較組

來源

  • Qin et al., 2018

    Tong Qin, Peiliang Li, Shaojie Shen（2018）VINS-Mono: A Robust and Versatile Monocular Visual-Inertial State EstimatorIEEE Transactions on Robotics, 34(4):1004-1020

    DOI 10.1109/tro.2018.2853729arXiv 1708.03852程式碼

    同儕審查已出版已讀全文近十年

程式碼：https://github.com/HKUST-Aerial-Robotics/VINS-Mono（授權：GPLv3 (README licence section)）。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