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

PlanarMesh

PlanarMesh 以「平面網格」表示場景：每個元素由一個平面（位置與法向量，以增量 PCA 更新）和落在該平面上的三角網格組成，頂點半徑近似局部曲率。每個新點經兩棵可動態插入的包圍體階層樹查詢：面相交搜尋（FIS）找出射線穿過的網格面以引入自由空間資訊，反向半徑搜尋（RRS）找出半徑涵蓋該點的邊界頂點；再以 95% 信賴區間的 z 檢定判定點在平面前方、平面上或後方，據以執行更新、擴張、新增、刪除、收縮與合併。輸出前依頂點半徑重新取樣並以 Delaunay 重新三角化，因此大平面用少量大三角形、細部用小三角形。位姿由外部 LiDAR 里程計提供。

地圖表示與表面重建近十年已讀全文同儕審查已出版地圖表示與表面重建

閱讀版本
arXiv 2510.13599v1 (8 pp., 15 Oct 2025, CC BY 4.0; the only arXiv version) read in full; the IEEE IROS 2025 version of record (pp. 15726-15733) was not compared in this pass
全文取得途徑
arXiv 預印本
本頁內容

An incremental LiDAR meshing system whose planar-mesh elements (a plane plus a mesh lying on it) are updated, grown, created, carved by free-space rays, shrunk and merged, with resolution set by per-vertex radii that track local curvature; poses come from an external odometry.

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not named in the paper; the text refers to walking speeds (a new scan every 0.5-1 m at normal walking speed; about 300 m at 1 m/s), which implies a person-carried scanner (inference)
狀態估計不適用 (poses from a separate LiDAR odometry such as FastLIO or VILENS; evaluation used ground-truth poses from registering each undistorted scan to the TLS map)
資料關聯per-point Face Intersection Search and Reverse Radius Search on bounding volume hierarchies to find candidate planar-meshes
時間表示不適用 (externally supplied poses)
去畸變delegated to the upstream LiDAR odometry (motion correction assumed in the supplied poses/scans)
迴圈閉合none (elastic deformation for future loop closure mentioned as future work)
全域最佳化none
地圖表示planar-mesh (plane models combined with adaptive-resolution mesh), stored with a bounding volume hierarchy
先驗資訊external poses
可輸出幾何compact triangle mesh saved as binary PLY with 280-550 K faces (about 10 MB) on the tested sequences; before output each planar-mesh is resampled by vertex radius and re-triangulated with Delaunay, restoring concavity by discarding faces outside the original (Sec. III-F, IV-C)
計算需求CPU only; about 0.4 s per scan on average (0.7 s peak), about 2 Hz, using all cores of a 28-core Intel i7 while baselines ran single-core at about 1 Hz (Sec. IV-A, IV-D)

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
LiDAR64-beam LiDAR (model not named)資料集感測器Oxford Spiresbeam count given only in the abstract, which ties the about 2 Hz rate to a 64-beam sensor; model not named; experiments used undistorted scans with ground-truth poses from registration to the TLS map (Sec. IV-A)(Wang et al., 2025a, Abstract; Sec. III-A; Sec. IV-A)
地面雷射掃描儀（TLS）TLS (instrument not named)參考或真值量測Oxford Spireshighly accurate TLS point cloud map; every undistorted scan registered to it to obtain ground-truth poses; also the accuracy reference(Wang et al., 2025a, Sec. IV-A; Fig. 6)
運算硬體28-core Intel i7 CPU執行運算平台未標示no GPU acceleration; PlanarMesh used all cores, baselines a single core(Wang et al., 2025a, Sec. IV-A)

論文圖片

只收錄原文以開放授權（open license）釋出的圖片，並依授權條件標示出處、圖號、授權與修改方式。

  • 以平面網格重建畫框、整個房間與內凹窗框，不同平面以不同顏色表示

    Fig. 1以平面網格重建畫框、整個房間與內凹窗框，不同平面以不同顏色表示

    出處：Wang et al., 2025a，Fig. 1。授權：CC BY 4.0。原始圖檔。修改：縮小至寬度不超過 1400 px，並轉存為 WebP 格式。

  • 四段場景中 VDBFusion、ImMesh、TLS 真值與 PlanarMesh 的重建比較，標出門框與凹窗等細節

    Fig. 6四段場景中 VDBFusion、ImMesh、TLS 真值與 PlanarMesh 的重建比較，標出門框與凹窗等細節

    出處：Wang et al., 2025a，Fig. 6。授權：CC BY 4.0。原始圖檔。修改：縮小至寬度不超過 1400 px，並轉存為 WebP 格式。

作者報告的優勢與限制

優勢

  • Accuracy on par with or better than ImMesh, VDBFusion and OctoMap on Oxford Spires against TLS ground truth, with meshes of 280-550 K faces versus 1.8-23 M and about 10 MB versus about 100 MB of accumulated LiDAR points (abstract, Sec. IV-B/C, Table I).

限制

  • Memory grows until it exceeds available memory at about 300 scans (about 300 m at 1 m/s walking); submapping is future work (Sec. V).
  • Real-time rate relies on all CPU cores while baselines were run on a single core, so runtime comparisons are not like-for-like (Sec. IV-A).
  • Meshes show small holes and rugged edges; smoothing or plane intersection left for future work (Sec. IV-B).
  • Deliberately biased toward precision over recall because tiny faces are dropped; recall 0.894 on Keble College 03 was the lowest of all methods (Sec. IV-B, Table I).
  • No loop closure; listed as future work (Sec. V).

營建工程相關證據

以 Oxford Spires 資料集的 Christ Church 03、Keble College 03 與 Observatory 01 三條室內外序列評估，真值為高精度 TLS 點雲地圖，屬既有建築而非施工中工地。作者指出建成環境以平面為主，並認為此表示可支援平面圖生成與 BIM，但本文沒有實作（Sec. III-B）。未在營建場域驗證。

原文驗證環境：公開基準、已完工建築、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 3 個比較組，合計 57 筆紀錄。

Wang et al., 2025a · Table II 本方法 28 筆

資料集與序列Oxford Spires · Christ Church 03

表格設定（擷取紀錄原文）：Ablation on Christ Church 03: number of scans for which seed planar-meshes are retained (0, 1, 10, All); All is the default setting; same metrics as Table I. Note: the text in Sec. IV-E attributes 0.295 s and 5.9 MB to 1-scan retention, whereas the table lists them for 0 (Wang et al., 2025a, Table II)

Time (s) per scan，Oxford Spires · Christ Church 03

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., 2025a 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：s；場景：existing building, walking survey

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

數值與出處
方法（原文寫法）報告值出處
PlanarMesh, seed planar-meshes kept for 0 scans本方法原文提出硬體：28-core Intel i7 CPU, no GPU; PlanarMesh uses all cores, baselines one core each (Sec. IV-A)0.295 s(Wang et al., 2025a, Table II)
PlanarMesh, seed planar-meshes kept for 1 scans本方法原文提出硬體：28-core Intel i7 CPU, no GPU; PlanarMesh uses all cores, baselines one core each (Sec. IV-A)0.347 s(Wang et al., 2025a, Table II)
PlanarMesh, seed planar-meshes kept for 10 scans本方法原文提出硬體：28-core Intel i7 CPU, no GPU; PlanarMesh uses all cores, baselines one core each (Sec. IV-A)0.384 s(Wang et al., 2025a, Table II)
PlanarMesh, seed planar-meshes kept for All scans本方法原文提出硬體：28-core Intel i7 CPU, no GPU; PlanarMesh uses all cores, baselines one core each (Sec. IV-A)0.313 s(Wang et al., 2025a, Table II)

Wang et al., 2025a · Table I 本方法 27 筆

表格設定（擷取紀錄原文）：Oxford Spires; each method meshes individual scans with ground-truth poses (every undistorted scan registered to the TLS map); meshes sampled to the raw scan point count; distances to the TLS map after pre-filtering areas not seen by both; precision, recall and F-score at 0.1 m; OctoMap voxel 0.05 m, ImMesh and VDBFusion 0.1 m, baselines configured for about 1 Hz on one core, PlanarMesh on all 28 cores; file size as PLY binary; OctoMap has no faces or vertices (N/A) (Wang et al., 2025a, Table I)

Per-Scan Time (s)，Oxford Spires · Christ Church 03 (about 307 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 Wang et al., 2025a 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：s；場景：existing buildings, indoor and outdoor (walking survey)

資料來源作者報告值(Wang et al., 2025a, Table I)

數值與出處
方法（原文寫法）報告值出處
VDBFusion硬體：28-core Intel i7 CPU, no GPU; PlanarMesh uses all cores, baselines one core each (Sec. IV-A)0.871 s(Wang et al., 2025a, Table I)
ImMesh硬體：28-core Intel i7 CPU, no GPU; PlanarMesh uses all cores, baselines one core each (Sec. IV-A)0.724 s(Wang et al., 2025a, Table I)
PlanarMesh (Ours)本方法原文提出硬體：28-core Intel i7 CPU, no GPU; PlanarMesh uses all cores, baselines one core each (Sec. IV-A)0.392 s(Wang et al., 2025a, Table I)
OctoMap硬體：28-core Intel i7 CPU, no GPU; PlanarMesh uses all cores, baselines one core each (Sec. IV-A)0.432 s(Wang et al., 2025a, Table I)

Wang et al., 2025a · Text Sec.IV-D 本方法 2 筆

資料集與序列Oxford Spires · Christ Church 03

表格設定（擷取紀錄原文）：Per-scan processing time of PlanarMesh on Christ Church 03 (Fig. 7 stacked plot summarised in text); values stated in text, not read off the plot (Wang et al., 2025a, Text Sec.IV-D)

average processing time per scan (about)，Oxford Spires · Christ Church 03

這張表在此指標與資料序列只列出本方法一筆，沒有可並列的其他方法，因此不畫圖，數值與出處見下表。這是 Wang et al., 2025a 在此表設定下報告的數值（author-reported results），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：s；場景：existing building, walking survey

數值與出處
方法（原文寫法）報告值出處
PlanarMesh本方法原文提出硬體：28-core Intel i7 CPU, no GPU; PlanarMesh uses all cores, baselines one core each (Sec. IV-A)0.4 s(Wang et al., 2025a, Sec. IV-D)

來源

  • Wang et al., 2025a

    Jiahao Wang, Nived Chebrolu, Yifu Tao, Lintong Zhang, Ayoung Kim, Maurice Fallon（2025）PlanarMesh: Building Compact 3D Meshes from LiDAR using Incremental Adaptive Resolution Reconstruction2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 15726-15733

    DOI 10.1109/iros60139.2025.11246204arXiv 2510.13599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/ori-drs/planar_mesh（授權：GPL-3.0 (repository LICENSE.txt)）。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