Semantic localization on BIM maps
作者將 BIM 依樓層拆分，經 IfcOpenShell 轉為網格後取樣成帶有構件類別的語意點雲地圖，免除事先以 SLAM 建圖。定位時先做點對面 ICP，再依語意一致性篩選並加權的 ICP 精化位姿。實驗在新加坡國立大學六層校舍（已完工使用）進行，參考軌跡取自離線 2D Cartographer SLAM，非獨立測量。
本頁內容
Localizes a 3D LiDAR on semantically labeled point maps sampled from BIM using coarse-to-fine semantic ICP, avoiding a prior SLAM map, validated in a completed university building.
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 (Velodyne VLP-16) |
|---|---|
| 原文測試平台 | 原文未報告: Fig. 5 shows a Velodyne VLP-16 with portable power and a laptop; the carrier is not stated; data collected with planar motion (Sec. 4.1, 4.4) |
| 狀態估計 | frame-to-map registration: point-to-plane ICP then semantic-weighted point-to-plane ICP with Huber kernel; previous pose as initial guess; no odometry (Alg. 2, Sec. 3.3) |
| 資料關聯 | coarse point-to-plane ICP, then each scan point is labelled only if all K nearest map points share one BIM category (Eq. 7); only floors, walls and columns are kept (chosen from the Seq. 3-1 test in Table 3); weighted point-to-plane ICP with w = w_c w_rho, semantic weight mu = 0.8 and Huber threshold delta = 0.05 m; about 3% of raw points reach the final step (Sec. 3.3, 4.3, Table 5) |
| 時間表示 | discrete poses |
| 去畸變 | 原文未報告 |
| 迴圈閉合 | none (localization only) |
| 全域最佳化 | none |
| 地圖表示 | storey-wise semantic point cloud sampled from BIM meshes (IfcOpenShell -> OBJ -> sampling at 30 points/m3 as written), labelled with the 13 BIM categories extracted with Dynamo, using non-oriented (axis-aligned) Dynamo bounding boxes searched with a k-d tree; about 40% walls, 20% floors and 20% curtain panels (Sec. 3.2, 4.1, 4.2) |
| 先驗資訊 | BIM (IFC) with element categories (Sec. 3.2) |
| 可輸出幾何 | pose trajectory in BIM frame (no new map) |
| 計算需求 | online localization on a low-power laptop, Intel i5-8265U, 16 GB RAM; libpointmatcher on ROS; mean time of Algorithm 2 per scan 108, 79, 112 and 114 ms on Seq. 2-2, 3-3, 4-1 and 5-1; authors state it can track a LiDAR operating at 10 Hz; maps prepared offline with Dynamo, CloudCompare and MATLAB (Sec. 4.1, 4.3) |
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 | Velodyne VLP-16 | 方法輸入 | 未標示 | single 3D LiDAR, no IMU or odometry used; operated at 10 Hz | (Yin et al., 2023, Sec. 4.1, Fig. 5, Sec. 4.3) |
| 運算硬體 | laptop with Intel I5-8265U, 16G RAM | 執行運算平台 | 未標示 | low-power laptop; libpointmatcher on ROS | (Yin et al., 2023, Sec. 4.1) |
作者報告的優勢與限制
優勢
- overall translation RMSE 0.080 m and rotation 0.663 deg; 34% translation improvement over baseline ICP (Table 4)
- semantic filtering alone (Sem (ORG)) improved overall translation error by 18% over ICP (ORG) (Sec. 4.3)
- Z drift (Delta Z) of the BIM-based localization stayed within 0.084 m on the four tested sequences, against up to -0.968 m (LOAM), -2.675 m (DLO) and -1.547 m (Open3D SLAM) (Table 6)
限制
- assumes static built environment; BIM updates needed for dynamic construction sites (Sec. 4.5)
- as-designed vs as-built deviations can cause failure (Sec. 4.5)
- depends on previous pose; no global localization; floor and initial pose set manually at the first frame (Sec. 4.5)
- reference trajectory from Cartographer SLAM run in 2D (x, y, yaw) with tuned parameters, not an external survey (Sec. 4.3)
- authors state that the BIM-based localization does not show better 2D pose accuracy than the LiDAR-only LOAM, DLO and Open3D SLAM on the four tested sequences; Table 6 is mixed: lowest translation RMSE on Seq. 3-3 but highest yaw RMSE on Seq. 2-3, 4-2 and 5-2 (Table 6, Sec. 4.4)
- axis-aligned Dynamo bounding boxes cause mixed or wrongly labelled map points (Fig. 8, Sec. 4.5)
- long, narrow corridors with only walls and floors degrade tracking and caused a trajectory discontinuity on the 4th storey (Sec. 4.5)
營建工程相關證據
測試於 NUS SDE4 已完工六層校舍的 2-5 樓（10 段、總長逾 340 m）；作者明言施工中動態環境需更新 BIM（Sec. 4.1, 4.5）。
原文驗證環境：已完工建築
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 38 筆紀錄。
Yin et al., 2023 · Table 4 本方法 22 筆
表格設定（擷取紀錄原文）：Ten self-collected VLP-16 sequences; RMSE of x-y translation and yaw against offline Cartographer 2D SLAM reference poses; trajectories aligned with the rpg trajectory-evaluation tool [50] (alignment type not stated); 40 ICP iterations for every method; columns ICP (w_rho) and Sem (w_c) omitted to respect the row cap (Yin et al., 2023, Table 4)
Tr. (m), RMSE of x-y translation error，self-collected NUS SDE4 sequences · 2-1
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。
按 Tab 進入圖表後，用上下方向鍵逐一瀏覽各類別，Esc 關閉提示框；也可開啟表格檢視閱讀全部數值。
這些是 Yin et al., 2023 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。
資料來源作者報告值(Yin et al., 2023, Table 4)
| 方法（原文寫法） | 報告值 | 出處 |
|---|---|---|
| ICP (ORG): libpointmatcher point-to-plane ICP with default geometric outlier filters (baseline) | 0.14 m | (Yin et al., 2023, Table 4) |
| Sem (ORG): ICP (ORG) plus semantic labelling and selection (ablation) | 0.127 m | (Yin et al., 2023, Table 4) |
| Sem (w_rho): semantic filtering with Huber weight only (ablation) | 0.115 m | (Yin et al., 2023, Table 4) |
| Sem (w_c w_rho): full semantic localization (proposed)本方法原文提出 | 0.113 m | (Yin et al., 2023, Table 4) |
Yin et al., 2023 · Table 6 本方法 12 筆
表格設定（擷取紀錄原文）：LiDAR-only comparison (no IMU for any method) on four sequences; 2D RMSE against the Cartographer 2D reference; Delta Z = mean height of the last 50 poses minus that of the first 50 poses, planar motion assumed (Yin et al., 2023, Table 6)
Tr. (m), 2D translation RMSE，self-collected NUS SDE4 sequences · 2-3
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。
按 Tab 進入圖表後，用上下方向鍵逐一瀏覽各類別，Esc 關閉提示框；也可開啟表格檢視閱讀全部數值。
這些是 Yin et al., 2023 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。
資料來源作者報告值(Yin et al., 2023, Table 6)
| 方法（原文寫法） | 報告值 | 出處 |
|---|---|---|
| LOAM [5] (A-LOAM code) | 0.058 m | (Yin et al., 2023, Table 6) |
| DLO [55] | 0.062 m | (Yin et al., 2023, Table 6) |
| Open3D SLAM [56] | 0.1 m | (Yin et al., 2023, Table 6) |
| BIM-based Localization (proposed)本方法原文提出 | 0.077 m | (Yin et al., 2023, Table 6) |
Yin et al., 2023 · Text Sec.4.3 本方法 4 筆
指標mean time cost of semantic localization (Algorithm 2)
表格設定（擷取紀錄原文）：Mean time cost of Algorithm 2 per scan in four case studies (Fig. 13, Table 5) (Yin et al., 2023, Text Sec.4.3)
mean time cost of semantic localization (Algorithm 2)，self-collected NUS SDE4 sequences · 2-2
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 Yin et al., 2023 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。
| 方法（原文寫法） | 報告值 | 出處 |
|---|---|---|
| Sem (w_c w_rho): full semantic localization (proposed)本方法原文提出硬體：low-power laptop, Intel i5-8265U, 16 GB RAM | 108 ms | (Yin et al., 2023, Sec. 4.3) |
來源
Yin et al., 2023
（2023）Semantic localization on BIM-generated maps using a 3D LiDAR sensorAutomation in Construction, 146, 104641
DOI 10.1016/j.autcon.2022.104641arXiv 2205.00816
同儕審查已出版已讀全文近十年
相關版本
- 預印本：Semantic localization on BIM-generated maps using a 3D LiDAR sensor (arXiv v1 2022-05-02, v2 2022-11-29) https://arxiv.org/abs/2205.00816