Point cloud to BIM registration (SLAM tracking)
作者以 Kudan LiDAR SLAM 追蹤 Ouster OS0-128 光達（含感測器 IMU 資料），將關鍵影格累積的點雲配準到以 IfcOpenShell 解析並體素化（0.1 m）的 BIM 點雲：先以法向量角度直方圖做軸向對齊，再以只保留垂直於投影面點的法向過濾樣板匹配（每 1° 測試）粗對位，最後以隨機子取樣 ICP 精對位，得到感測器在 BIM 座標中的位姿，供擴增實境檢查與 Boston Dynamics Spot 遠端操作使用。評估只用手持錄製資料，且以每段錄製起點人工量測的初始位置作為唯一參考：28 m 走廊（只含走廊的 BIM）各關鍵影格中位數平均 XY 誤差 0.03 m、Z 0.035 m，全部成功；TU Wien 圖書館六樓含多個相似房間的環形走廊，前 10 個關鍵影格成功率只有 30%，到第 36 個關鍵影格前平均 XY 0.19 m、Z 0.24 m，之後 XY 低於 0.3 m、Z 低於 0.4 m。誤差同時包含配準誤差與 SLAM 漂移。
本頁內容
Registers a Kudan-SLAM keyframe map from an Ouster OS0-128 to a voxelized IFC model (axis alignment, normal-filtered template matching, ICP); against hand-measured start positions it reaches 0.03 m (median XY) with a hallway-only BIM, but on a self-similar library floor succeeds only 30% of the time in the first keyframes and averages 0.19 m XY and 0.24 m Z, with drift pushing errors toward 0.3-0.4 m.
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 | Ouster OS0-128 Gen 2 LiDAR (128 channels, 90 deg VFOV; 512x20 in env. 1, 1024x20 in env. 2; up to 131,072 points per frame) with IMU data from the sensor |
|---|---|
| 原文測試平台 | handheld: custom wooden frame with dual hold and 24 V battery (all evaluation recordings)、legged robot Boston Dynamics Spot (demonstration setup, not evaluated) |
| 狀態估計 | Kudan LiDAR SLAM (commercial SDK; tracking voxel 0.25 m) for pose tracking; point cloud to BIM registration of the accumulated keyframe map gives the SLAM-to-BIM transform, repeatable when drift grows |
| 資料關聯 | PCA normals and normal-angle histograms for axis alignment; normal-filtered template matching (points kept only when the dot product of point normal and plane normal is at most 0.1): rotation from XY projections tested at each 1 deg, first at lower resolution and then at the 0.1 m resolution, height matched with projections along the X and Y axes; ICP with random sub-sampling, 100 iterations split into 10 random 10% subsets |
| 時間表示 | 原文未報告 |
| 去畸變 | 原文未報告 |
| 迴圈閉合 | may occur inside Kudan SLAM, but neither loop-closure accuracy nor its influence on registration was evaluated |
| 全域最佳化 | 未查證 |
| 地圖表示 | keyframe-accumulated LiDAR point cloud from Kudan SLAM; BIM converted to a voxel point cloud (0.1 m) with IfcOpenShell and the Voxelization Toolkit for one manually selected floor (env. 2 floor: 1,035,690 points) |
| 先驗資訊 | BIM (IFC) of the relevant floor, selected manually; hallway BIM derived from a high-resolution scan, library BIM from 2D floor plans and facade photogrammetry |
| 可輸出幾何 | sensor pose in BIM frame |
| 計算需求 | handheld: all components on a Razer Blade 15 laptop (11th Gen Intel Core i7, 8 cores, 32 GB RAM, NVIDIA GeForce RTX 3080, Windows 11); robot: tracking on an Intel NUC (Core i7, 4 cores, 16 GB RAM, Windows 10), localization and visualization on the laptop over WiFi; registration time split 2% axis alignment, 58% template matching, 40% ICP |
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 | Ouster OS0-128 Gen 2歸入：Ouster OS0-128 | 方法輸入 | 未標示 | 128 channels, 90 deg vertical FOV; resolution 512x20 (env. 1) or 1024x20 (env. 2); up to 131,072 points per frame; raw data recorded as .pcap with Ouster Studio | (Schaub et al., 2022, Sec. 3.3, 4.1, 4.3) |
| 慣性量測單元（IMU） | IMU of the Ouster OS0-128 (model not reported) | 方法輸入 | 未標示 | IMU data from the LiDAR sensor fed to SLAM tracking | (Schaub et al., 2022, Sec. 3.1) |
| 載具平台 | custom-built wooden handheld frame with dual hold | 方法輸入 | 未標示 | 24 V battery; Ethernet to laptop; used for all evaluation recordings | (Schaub et al., 2022, Sec. 3.3, Fig. 5) |
| 載具平台 | Boston Dynamics Spot | 方法輸入 | 未標示 | LiDAR and payload computer powered by the robot; demonstration only | (Schaub et al., 2022, Sec. 3.3, Figs. 1, 5) |
| 運算硬體 | Intel NUC (payload computer on Spot) | 執行運算平台 | 未標示 | Intel Core i7 CPU (4 cores), 16 GB RAM, Windows 10; runs the real-time data and tracking component | (Schaub et al., 2022, Sec. 3.3) |
| 運算硬體 | Razer Blade 15 | 執行運算平台 | 未標示 | 11th Gen Intel Core i7 (8 cores), 32 GB RAM, NVIDIA GeForce RTX 3080, Windows 11; runs all components in the handheld setup and localization plus visualization in the robot setup | (Schaub et al., 2022, Sec. 3.3) |
作者報告的優勢與限制
優勢
- env. 1 (28 m x 2.5 m hallway, hallway-only BIM): XY error 0.03 m and Z 0.035 m (medians averaged over keyframes); registration successful at all keyframes of 10 recordings (Sec. 4.3)
- env. 2: registration virtually always successful from keyframe 28; mean XY 0.19 m and Z 0.24 m up to keyframe 36 (Sec. 4.3)
- supports non-perpendicular walls and varying floor levels, unlike Herbers and Konig (Sec. 5)
- registration runs in the background without degrading live tracking (Sec. 4.3)
限制
- self-similar floor plan causes wrong template matches for up to 20 first keyframes; success only 30% in keyframes 2-10 and 50-90% in keyframes 12-26 (Sec. 4.3, 5)
- error grows with accumulated SLAM drift, notably after keyframe 38 (about 60.8 m); afterwards XY below 0.3 m and Z below 0.4 m (Sec. 4.3)
- ground truth only for the initial sensor position, measured manually; SLAM drift contribution not quantified (Sec. 4.1)
- differences between BIM and building (furniture, plants, missing or extra walls, doors, windows) hamper registration (Sec. 5)
- only corridors on one floor recorded; rooms not accessible (Sec. 4.2)
營建工程相關證據
應用動機為建築進度控制、數位輔助維護與遠端巡檢；測試在 TU Wien 研究室旁約 28 m 長、2.5 m 寬的走廊，以及 TU Wien 圖書館六樓的環形走廊進行，兩者都是既有建築；圖書館樓層可見 BIM 未記載的家具、植物與牆體、門窗差異，且無法進入個別房間；非施工中工地（Sec. 1、4.2、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 2 個比較組，合計 9 筆紀錄。
Schaub et al., 2022 · Text Sec.4.3 本方法 8 筆
資料集與序列TU Wien hallway (test environment 1) · 10 recordings, keyframes 1-9 (mean 21.5 m travelled at keyframe 9)
表格設定（擷取紀錄原文）：Localization error = registration-derived initial sensor position vs manually measured initial position in the BIM frame, evaluated at every second keyframe; env. 1 reports medians averaged over keyframes; 10 handheld recordings, 512x20. (Schaub et al., 2022, Text Sec.4.3)
XY error averaged over all keyframes，TU Wien hallway (test environment 1) · 10 recordings, keyframes 1-9 (mean 21.5 m travelled at keyframe 9)
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 Schaub et al., 2022 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 |
|---|---|---|
| Kudan SLAM + point cloud to BIM registration (proposed)本方法原文提出 | 0.03 m | (Schaub et al., 2022, Sec. 4.3, Fig. 9) |
Schaub et al., 2022 · Text Sec.5 本方法 1 筆
指標average Z-axis accuracy at the best interval
資料集與序列TU Wien library, 6th floor (test environment 2) · keyframes 20-38
表格設定（擷取紀錄原文）：Best interval in env. 2: after keyframe 20 (about 25 m) and before keyframe 38 (about 60 m); average values. (Schaub et al., 2022, Text Sec.5)
average Z-axis accuracy at the best interval，TU Wien library, 6th floor (test environment 2) · keyframes 20-38
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 Schaub et al., 2022 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 |
|---|---|---|
| Kudan SLAM + point cloud to BIM registration (proposed)本方法原文提出 | 0.25 m | (Schaub et al., 2022, Sec. 5) |
來源
Schaub et al., 2022
（2022）Point cloud to BIM registration for robot localization and Augmented Reality2022 IEEE International Symposium on Mixed and Augmented Reality Adjunct (ISMAR-Adjunct), 77-84
DOI 10.1109/ismar-adjunct57072.2022.00025
同儕審查已出版已讀全文近十年