SLAM 點雲圖鑑Point clouds for construction
M-01 · 2020

Removert

Removert 以多解析度距離影像（range image）比較查詢掃描與含動態點的累積地圖：先保守地只保留確定的靜態點，再逐步放大查詢與地圖的關聯視窗，把被誤刪的靜態點「回復」（revert），藉此隱式補償位姿估計與配準誤差。方法離線處理，輸入為任一光達里程計或 SLAM 輸出的掃描與位姿。

地圖表示與表面重建近十年已讀全文同儕審查已出版全域一致性、場所辨識與長期建圖

閱讀版本
Version of record, IEEE Xplore HTML full text (IROS 2020, document 9340856)
全文取得途徑
國立臺灣大學機構訂閱
本頁內容

Removert removes dynamic points by multi-resolution range-image comparison between scans and the accumulated map, then reverts falsely removed static points by enlarging the association window to tolerate pose errors.

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
原文測試平台vehicle (KITTI 01, 03, 08, 09 with SemanticKITTI labels; MulRan KAIST 02 qualitative)
狀態估計不適用
資料關聯visibility check in range images: the map (per query frame, using its SE(3) pose) and the query scan are projected to fixed-resolution range images keeping the minimum range per pixel; a map point is marked dynamic when the query-minus-map range difference exceeds a range-adaptive threshold (tau_D times range); batch voting over N randomly ordered scans gives a staticity score (alpha_SM = 0.3, alpha_DM = -0.7, tau_S = -0.1); removal at the finest resolution (vertical FOV / number of rays, 0.4 deg on KITTI) for three steps, then seven revert iterations coarsening by 0.1 deg per iteration
時間表示不適用
去畸變原文未報告
迴圈閉合不適用
全域最佳化不適用
地圖表示static point-cloud map (dynamic points separated)
先驗資訊raw scans with SE(3) poses from LiDAR SLAM assumed to contain some error; experiments deliberately used SuMa poses (as in SemanticKITTI) rather than KITTI ground truth; a batch of sequential scans (e.g. 50 scans for a 100 m submap at 2 m spacing)
可輸出幾何static point-cloud map and parsed dynamic points (README)
計算需求offline batch post-processing; the paper explicitly sets processing speed aside and reports no runtime or hardware (the over 10 Hz figure comes from the code README only)

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
LiDAR原文未報告 (KITTI LiDAR, 64 rays, 27 deg vertical FOV)資料集感測器KITTI odometry; SemanticKITTIvertical FOV 27 deg, 64 rays(Kim & Kim, 2020, Sec. IV-A)
LiDAR原文未報告 (MulRan LiDAR, 45 deg vertical FOV)資料集感測器MulRan (KAIST 02)vertical FOV 45 deg(Kim & Kim, 2020, Sec. IV-B, Fig. 7)

作者報告的優勢與限制

優勢

  • Tolerates imperfect pose estimation via multiresolution revert step, shown with SuMa poses instead of ground truth (Sec. IV-A)
  • Authors report competing with or exceeding human labels in ambiguous regions on KITTI, e.g. tree leaves left unlabeled in SemanticKITTI (Sec. IV-B, Fig. 6)
  • Needs only point coordinates, without normals, incidence angles or region growing (Sec. III-D)
  • Worked qualitatively for LiDARs with 27 deg (KITTI) and 45 deg (MulRan) vertical FOV (Sec. IV-B, Fig. 7)
  • By-product: self-labelled dynamic objects per scan by nearest-point checks against the static map (Sec. IV-C, Fig. 10)

限制

  • Offline only: the paper excludes online operation and does not report speed (Sec. II-A)
  • Authors' stated limitation: the map range image keeps only the nearest visible point per pixel, so dynamic points behind static structures such as a concrete median barrier are not removed (Sec. IV-D, Fig. 11)
  • Quantitative evaluation limited to TP, FP and FN point counts on one 100 m KITTI 03 segment, with no numeric comparison against other removal methods (Sec. IV-B, Fig. 9)
  • Too coarse a resolution wrongly recovers points near the ground, e.g. under cars (Fig. 3 caption)
  • Non-commercial license (README)
  • Found in follow-up benchmark: under point-wise evaluation Removert retained almost all static points but labelled few dynamic points; misses are linked to occlusion behind true dynamic points (Zhang et al., 2023a Sec. V-A, Table I)

營建工程相關證據

原文未報告；被 LT-mapper 用於高動態點移除與變化偵測（Kim & Kim, 2022 Sec. V-A1）。

原文驗證環境：公開基準

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 9 個比較組，合計 90 筆紀錄。以下列出本方法紀錄最多的 4 組，其餘 5 組列在最後，並連到性能比較頁。

Lim et al., 2021 · Table II 本方法 30 筆

表格設定（擷取紀錄原文）：Static-map benchmark on five manually selected SemanticKITTI frame ranges with SuMa poses; PR and RR computed voxel-wise with 0.2 voxel size for all methods; OctoMap run at 0.05 and 0.2 voxel sizes; Removert RM3 = three removal stages, RM3+RV1 adds one revert stage (Lim et al., 2021, Table II)

Preservation Rate (PR)，SemanticKITTI · 00 (frames 4390-4530)

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 Lim 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：未對齊；單位：%；場景：urban driving (countryside, highway, intersections)

資料來源作者報告值(Lim et al., 2021, Table II)

數值與出處
方法（原文寫法）報告值出處
OctoMap - 0.0576.731%(Lim et al., 2021, Table II)
OctoMap - 0.234.568%(Lim et al., 2021, Table II)
Peopleremover37.523%(Lim et al., 2021, Table II)
Removert - RM3本方法85.502%(Lim et al., 2021, Table II)
Removert - RM3+RV1本方法86.829%(Lim et al., 2021, Table II)
ERASOR (Ours)原文提出93.98%(Lim et al., 2021, Table II)

Yang et al., 2024 · Table II 本方法 18 筆

表格設定（擷取紀錄原文）：Dynamic object removal on SemanticKITTI; point-wise labels, moving classes counted as dynamic; sequences and scan ranges follow the ERASOR setup; authors note ERASOR ran at a lower frame rate; baseline execution settings otherwise not stated (Yang et al., 2024, Table II)

PR (preservation rate)，SemanticKITTI · 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 Yang et al., 2024 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：未對齊；單位：ratio；場景：urban driving, vehicle-mounted LiDAR

資料來源作者報告值(Yang et al., 2024, Table II)

數值與出處
方法（原文寫法）報告值出處
ERASOR0.9172 ratio(Yang et al., 2024, Table II)
Removert本方法0.9328 ratio(Yang et al., 2024, Table II)
Ground-Octomap0.7765 ratio(Yang et al., 2024, Table II)
Ours原文提出0.9471 ratio(Yang et al., 2024, Table II)

Duberg et al., 2024 · Table I 本方法 12 筆

表格設定（擷取紀錄原文）：Point-wise dynamic point removal accuracy (%) following the DynamicMap benchmark protocol; Removert, ERASOR, OctoMap and DUFOMap evaluated offline, Dynablox and DUFOMap* online (each scan classified with the map built so far); DUFOMap uses the same parameters for all data (voxel 0.1 m, ds 0.2 m, dp 1), Removert and ERASOR per-dataset optimized parameters; KITTI labels and poses from SemanticKITTI (Duberg et al., 2024, Table I)

SA (static accuracy, share of static points correctly kept)，KITTI (SemanticKITTI labels and poses) · 00 small town

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 Duberg et al., 2024 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：未對齊；單位：%；場景：small town (HDL-64E)

資料來源作者報告值(Duberg et al., 2024, Table I)

數值與出處
方法（原文寫法）報告值出處
Removert [8]本方法99.44%(Duberg et al., 2024, Table I)
ERASOR [9]66.7%(Duberg et al., 2024, Table I)
OctoMap [16]68.05%(Duberg et al., 2024, Table I)
DUFOMap (Ours)原文提出97.96%(Duberg et al., 2024, Table I)
Dynablox [17]96.76%(Duberg et al., 2024, Table I)
DUFOMap* (Ours, online)原文提出98.37%(Duberg et al., 2024, Table I)

Zhang et al., 2023a · Table I 本方法 12 筆

表格設定（擷取紀錄原文）：Point-wise dynamic point removal accuracy (%): SA static accuracy, DA dynamic accuracy, AA = sqrt(SA x DA); methods marked * are offline and need a prior raw map; 'Octomap w G' adds ground estimation and 'Octomap w GF' also statistical outlier filtering (the benchmark's own extension); poses from dataset files for KITTI and AV2.0, simple NDT SLAM for semi-indoor (Zhang et al., 2023a, Table I)

SA (static accuracy)，KITTI (SemanticKITTI labels) · sequence 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 Zhang et al., 2023a 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：未對齊；單位：%；場景：small town (SemanticKITTI)

資料來源作者報告值(Zhang et al., 2023a, Table I)

數值與出處
方法（原文寫法）報告值出處
Removert* [5]本方法99.44%(Zhang et al., 2023a, Table I)
ERASOR* [16]66.7%(Zhang et al., 2023a, Table I)
Octomap [8]68.05%(Zhang et al., 2023a, Table I)
Octomap w G原文提出85.92%(Zhang et al., 2023a, Table I)
Octomap w GF原文提出93.06%(Zhang et al., 2023a, Table I)

其他比較組

列出其餘 5 個比較組

來源

  • Kim & Kim, 2020

    Giseop Kim, Ayoung Kim（2020）Remove, then Revert: Static Point cloud Map Construction using Multiresolution Range Images2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 10758-10765

    DOI 10.1109/iros45743.2020.9340856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/gisbi-kim/removert（授權：CC BY-NC-SA 4.0 (stated in README; non-commercial)）。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