SLAM 點雲圖鑑Point clouds for construction
M-01 · 2017

LOAM (journal version)

本記錄為 LOAM 的期刊版本（Autonomous Robots，2016-02-18 線上發表、2017 年卷期）。方法核心與 RSS 2014 版相同：高頻、低精度的里程計估計速度並去除點雲運動畸變，低頻（預設為里程計的十分之一）的建圖以更多特徵點精細配準；兩者都使用依局部平滑度挑出的邊緣點與平面點，以 Levenberg-Marquardt 搭配 bisquare 權重求解，IMU 僅為選用的前處理先驗，且不含迴圈閉合。相較 RSS 版，期刊版把驗證擴充到四種感測系統（含八旋翼機與 Velodyne HDL-32E 車載測試），建圖以 5 cm（邊緣）與 10 cm（平面）體素平均，並說明 KITTI 測試為求精度改為逐幀建圖、只達即時速度的約 10%。

里程計＋局部建圖經典已讀全文同儕審查已出版LOAM 系列與特徵式 LiDAR 建圖

閱讀版本
version of record: Springer publisher PDF, Auton Robot (2017) 41:401-416, published online 2016-02-18 (16 pages)
全文取得途徑
國立臺灣大學機構訂閱
本頁內容

Journal version of LOAM: same odometry-mapping split with smoothness-selected edge and planar features and robust LM, no loop closure; adds octo-rotor and HDL-32E vehicle tests, separate 5 cm and 10 cm voxel sizes for edge and planar maps, and notes that its KITTI runs used per-scan mapping at 10% of real time.

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2-axis lidar: back-and-forth spinning Hokuyo UTM-30LX (Sec. 4.1)、continuously spinning Hokuyo on an octo-rotor (Sec. 7.3)、Velodyne HDL-32E (Sec. 7.4)、Velodyne HDL-64E via KITTI (Sec. 7.5)、IMU optional: Xsens MTi-10 (Sec. 7.2), Microstrain 3DM-GX3-45 (Sec. 7.3)
原文測試平台cart (pushed, indoor)、ground vehicle、handheld、octo-rotor micro aerial vehicle、utility vehicle and passenger vehicle、vehicle (KITTI)
狀態估計Levenberg-Marquardt adapted to robust fitting with bisquare weights; odometry at scan rate and mapping at about one-tenth of that rate (ratio 10 preferred, Sec. 8); for KITTI the mapping ran every scan, giving 10% of real-time speed (Sec. 7.5)
資料關聯edge and planar feature points selected by local smoothness; point-to-edge-line and point-to-planar-patch distances; KD-tree nearest neighbours (Sec. 5-6)
時間表示constant angular and linear velocity within a sweep, linear pose interpolation (Sec. 3-5)
去畸變odometry-estimated velocity used to reproject points; optional IMU pre-processing removes orientation change and part of the acceleration effect (Sec. 7.2)
迴圈閉合none; authors state they do not consider loop closure (Sec. 1) and list it as future work (Sec. 9)
全域最佳化none
地圖表示feature point cloud map stored in 10 m cubes with KD-tree search; voxel-grid averaging at 5 cm (edge) and 10 cm (planar); map truncated to a 500 m cube around the sensor (Sec. 6)
先驗資訊none (IMU optional)
可輸出幾何registered feature point cloud map and 6-DoF pose (Sec. 6); dense raw-point export not described in the text read
計算需求Laptop with 2.5 GHz quad cores and 6 GiB memory, ROS on Linux, odometry and mapping on two separate threads (Sec. 7). Per-call totals: Hokuyo accuracy tests odometry 48 ms and mapping 309 ms (Table 1); Velodyne HDL-32E tests odometry 69 ms and mapping 563 ms (Table 4). For KITTI, mapping was run on every scan, about 1 s per scan, i.e., 10% of real-time speed (Sec. 7.5).

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
LiDARHokuyo UTM-30LX方法輸入未標示2D scanner turned into a 2-axis back-and-forth spinning lidar by a motor and an encoder (Fig. 2); 180 deg FoV, 0.25 deg resolution, 40 lines/s; motor rotates at 180 deg/s between -90 and 90 deg, one sweep lasts 1 s; onboard encoder with 0.25 deg resolution(Zhang & Singh, 2017, Sec. 4.1, Fig. 2)
LiDARHokuyo laser scanner方法輸入未標示scanner model not stated; continuously spinning 2-axis lidar of the same design as Fig. 2; sweep is a semi-spherical rotation on the slow axis lasting 1 s(Zhang & Singh, 2017, Sec. 7.3, Fig. 14)
LiDARVelodyne HDL-32E方法輸入未標示single-axis scanner with 32 beams, 10 Hz by default; mounted high on vehicle roof(Zhang & Singh, 2017, Sec. 7.4, Fig. 17)
LiDARVelodyne HDL-64E資料集感測器KITTI odometrylogged at 10 Hz(Zhang & Singh, 2017, Sec. 7.5, Fig. 20)
慣性量測單元（IMU）Xsens MTi-10方法輸入未標示orientation from gyros and accelerometers fused in a Kalman filter; used to pre-process the point cloud(Zhang & Singh, 2017, Sec. 7.2)
慣性量測單元（IMU）Microstrain 3DM-GX3-45方法輸入未標示原文未報告(Zhang & Singh, 2017, Sec. 7.3)
GNSS 接收器KITTI high accuracy GPS/INS (model not stated)參考或真值量測KITTI odometryground truth for sequences 0-10(Zhang & Singh, 2017, Sec. 7.5)
GNSS 接收器High accuracy GPS/INS on the ground vehicle (model not stated)參考或真值量測未標示ground truth for orchard drift tests(Zhang & Singh, 2017, Sec. 7.1, Table 2)
雙目相機KITTI color and monochrome stereo cameras (model not stated)資料集感測器KITTI odometrynot used by the method(Zhang & Singh, 2017, Sec. 7.5, Fig. 20)
載具平台Pushed cart (indoor)方法輸入未標示carries lidar, battery and laptop; pushed by a walking person at 0.5 m/s(Zhang & Singh, 2017, Sec. 7.1, Fig. 10)
載具平台Ground vehicle (outdoor)方法輸入未標示lidar mounted at the front; 0.5 m/s(Zhang & Singh, 2017, Sec. 7.1, Fig. 10)
載具平台Handheld (person holding the lidar)方法輸入未標示walking at 0.5 m/s while moving the lidar up and down about 0.5 m; staircase test(Zhang & Singh, 2017, Sec. 7.2, Fig. 13)
載具平台Octo-rotor micro aerial vehicle方法輸入未標示manually flown at 1 m/s(Zhang & Singh, 2017, Sec. 7.3, Fig. 14)
載具平台Utility vehicle方法輸入未標示sidewalks and off-road terrain; 2-3 m/s on 1.0 km campus run(Zhang & Singh, 2017, Sec. 7.4, Fig. 17a)
載具平台Passenger vehicle方法輸入未標示streets; mostly 11-18 m/s on 3.6 km run(Zhang & Singh, 2017, Sec. 7.4, Fig. 17b)
運算硬體Laptop computer (model not stated)執行運算平台未標示2.5 GHz quad cores, 6 GiB memory, ROS on Linux; odometry and mapping on two threads(Zhang & Singh, 2017, Sec. 7)
其他Tape ruler歸入：tape ruler參考或真值量測未標示manual ground truth for handheld tests(Zhang & Singh, 2017, Sec. 7.2, Table 3)
其他Satellite image參考或真值量測未標示trajectory and building walls matched to it to judge horizontal drift(Zhang & Singh, 2017, Sec. 7.4, Figs. 18-19)

作者報告的優勢與限制

優勢

  • KITTI odometry benchmark: 0.88% average position error, ranked #2, and reported to outperform stereo visual odometry methods by over 14% in position and 24% in orientation error (Sec. 7.5)
  • Hokuyo drift tests at 0.5 m/s: corridor 0.9% (58 m) and 1.1% (46 m) from the start and end gap of a closed loop, orchard 2.3% (52 m) and 2.8% (67 m) against GPS/INS (Table 2)
  • handheld tests with tape-ruler ground truth: IMU pre-processing plus the method gave the lowest error in all four scenes, e.g., corridor 0.9% versus 2.1% without IMU and 16.7% with IMU orientation only (Table 3)
  • HDL-32E campus run 1.0 km: horizontal drift <=1 m, vertical drift <=1.5 m, overall <=0.2% of distance
  • street run 3.6 km: horizontal error <=2 m, both from satellite-image matching (Sec. 7.4)

限制

  • No loop closure
  • correcting drift by closing loops is left to future work (Sec. 1, 9)
  • KITTI accuracy was obtained with mapping on every scan, so the system ran at 10% of real-time speed (Sec. 7.5)
  • assumes smooth, continuous angular and linear velocity within a sweep unless an IMU pre-processes the data (Sec. 3, 7.2)
  • matching errors were larger in natural outdoor scenes than in indoor scenes (Sec. 7.1, Fig. 11)
  • octo-rotor flights had no ground truth and were judged only visually (Sec. 7.3)
  • vertical accuracy of the 3.6 km street run could not be evaluated (Sec. 7.4)
  • follow-up statements retained: LeGO-LOAM Sec. IV-D (10% real time) and LIO-SAM ref [1] (IMU used only for de-skewing and motion prior)

營建工程相關證據

原文未報告（作者在引言主張在許多實務情況，例如建築物單一樓層的建圖，迴圈閉合並非必要；此為作者主張，論文未在營建工地或以工程幾何參考驗證，室內測試僅為既有建築走廊與大廳）。

原文驗證環境：公開基準、受控實驗、已完工建築、獨立參考量測、跨場域

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 28 個比較組，合計 200 筆紀錄。以下列出本方法紀錄最多的 4 組，其餘 24 組列在最後，並連到性能比較頁。

Li et al., 2019 · Table 1 本方法 22 筆

表格設定（擷取紀錄原文）：KITTI odometry metric: t_rel = average translational RMSE (%) and r_rel = average rotational RMSE (deg/100 m) over 100-800 m lengths. LO-Net trained on KITTI 00-06 and tested on 07-10 and on Ford without fine-tuning; loop closure disabled for all methods. LOAM values outside brackets come from the authors' modified re-run; bracketed values are quoted from the LOAM paper [45]. Velas et al. values quoted from [35] (r_rel and Ford NA). ICP variants run with PCL. Truncated: per-sequence rows 00-06 (training sequences) omitted; the mean over them (mean-dagger) is kept. (Li et al., 2019, Table 1)

t_rel: average translational RMSE (%) on length of 100 m-800 m，KITTI odometry · 07 (not used for training)

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 Li et al., 2019 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：平均值（mean）；對齊方式：未對齊；單位：%；場景：urban, country and highway driving, vehicle-mounted Velodyne HDL-64

資料來源作者報告值(Li et al., 2019, Table 1)

數值與出處
方法（原文寫法）報告值出處
ICP-po2po (PCL)5.17%(Li et al., 2019, Table 1)
ICP-po2pl (PCL)1.55%(Li et al., 2019, Table 1)
GICP [30]0.64%(Li et al., 2019, Table 1)
CLS [34]1.04%(Li et al., 2019, Table 1)
LOAM [45] (authors' modified re-run)本方法0.69%(Li et al., 2019, Table 1)
Velas et al. [35] (values from [35])1.77%(Li et al., 2019, Table 1)
LO-Net原文提出1.7%(Li et al., 2019, Table 1)
LO-Net+Mapping原文提出0.56%(Li et al., 2019, Table 1)

Wang et al., 2021c · Table 1 本方法 13 筆

表格設定（擷取紀錄原文）：KITTI odometry, trained on 00-06 (marked *) and tested on 07-10; trel = average translational RMSE (%) over 100-800 m subsequences; rows other than LOAM w/o mapping and Ours are copied from LO-Net [10]; LOAM is a full system with mapping, others are odometry only (Wang et al., 2021c, Table 1)

trel (average translational RMSE, %)，KITTI odometry · 07 (test)

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 Wang et al., 2021c 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：不適用；單位：%；場景：vehicle, road

資料來源作者報告值(Wang et al., 2021c, Table 1)

數值與出處
方法（原文寫法）報告值出處
Full LOAM [31]本方法0.69%(Wang et al., 2021c, Table 1)
ICP-po2po5.17%(Wang et al., 2021c, Table 1)
ICP-po2pl1.55%(Wang et al., 2021c, Table 1)
GICP [19]0.64%(Wang et al., 2021c, Table 1)
CLS [21]1.04%(Wang et al., 2021c, Table 1)
Velas et al. [22]1.77%(Wang et al., 2021c, Table 1)
LO-Net [10]1.7%(Wang et al., 2021c, Table 1)
DMLO [11]0.73%(Wang et al., 2021c, Table 1)
LOAM w/o mapping (published code run by authors)10.87%(Wang et al., 2021c, Table 1)
Ours (PWCLO-Net)原文提出0.6%(Wang et al., 2021c, Table 1)

Zhang & Singh, 2017 · Table 3 本方法 12 筆

指標motion estimation error relative to distance (tape-ruler ground truth)

表格設定（擷取紀錄原文）：Motion estimation errors with and without IMU; handheld lidar, 0.5 m/s walking, ground truth by tape ruler (Zhang & Singh, 2017, Table 3)

motion estimation error relative to distance (tape-ruler ground truth)，author-collected handheld Hokuyo datasets · Corridor (32 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 Zhang & Singh, 2017 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：%；場景：corridor

資料來源作者報告值(Zhang & Singh, 2017, Table 3)

數值與出處
方法（原文寫法）報告值出處
IMU (orientation from IMU only, translation from the method)本方法16.7%(Zhang & Singh, 2017, Table 3)
Ours (no IMU)本方法原文提出2.1%(Zhang & Singh, 2017, Table 3)
Ours+IMU (IMU pre-processing followed by the method)本方法原文提出0.9%(Zhang & Singh, 2017, Table 3)

Li et al., 2021a · Table II 本方法 12 筆

表格設定（擷取紀錄原文）：KITTI odometry 00-10; mean relative pose error over 100-800 m trajectories (rotation deg/100m / translation %); * marks sequences with loops; LOAM values quoted from its journal paper [19]; other baselines run with open-source code (Li et al., 2021a, Table II)

relative translational error (%)，KITTI odometry · 00*

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 Li et al., 2021a 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：平均值（mean）；對齊方式：不適用；單位：%；場景：vehicle, road

資料來源作者報告值(Li et al., 2021a, Table II)

數值與出處
方法（原文寫法）報告值出處
LOAM* (from [19])本方法0.78%(Li et al., 2021a, Table II)
FLOAM0.92%(Li et al., 2021a, Table II)
ISC-LOAM1.02%(Li et al., 2021a, Table II)
SUMA0.77%(Li et al., 2021a, Table II)
SUMA++0.65%(Li et al., 2021a, Table II)
Ours-ODOM原文提出0.59%(Li et al., 2021a, Table II)
Ours-LOOP原文提出0.59%(Li et al., 2021a, Table II)

其他比較組

列出其餘 24 個比較組

來源

  • Zhang & Singh, 2017

    Ji Zhang, Sanjiv Singh（2017）Low-drift and real-time lidar odometry and mappingAutonomous Robots, 41(2): 401-416

    DOI 10.1007/s10514-016-9548-2

    同儕審查已出版已讀全文經典查證後修正

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