OctoMap
OctoMap 以八元樹（octree）儲存體素的機率佔據值（log-odds），同時表示已佔據、空與未知空間；感測器原點到端點之間的射線更新為空，端點更新為佔據。作者加入機率上下限夾制（clamping）與樹狀剪枝壓縮，使地圖精簡並可做多解析度查詢。
本頁內容
Probabilistic occupancy mapping in an octree that represents occupied, free and unknown space with clamped log-odds updates and lossless compression.
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 | ["2D laser range finder on a pan-tilt unit (SICK LMS, FR-079 corridor)", "two fixed laser scanners sweeping to the left and right of the robot (New College Epoch C、models not stated)", "dense 3D laser scans (Freiburg campus, 81 scans、sensor model not stated、ranges up to 50 m)", "RGB-D camera (Microsoft Kinect, freiburg1_360 sequence)"] |
|---|---|
| 原文測試平台 | ["wheeled robot (Pioneer2 AT, FR-079 corridor)", "hand-held Microsoft Kinect (freiburg1_360, TUM RGB-D)", "robot carrying two fixed sweeping laser scanners (New College Epoch C、platform type not stated in the paper)", "Freiburg campus dataset (platform not stated in the paper)"] |
| 狀態估計 | 不適用 (mapping with externally supplied poses: FR-079 odometry refined by 3D scan matching, New College trajectory from visual odometry (Sibley et al. 2009), freiburg1_360 aligned by RGB-D SLAM) |
| 資料關聯 | ray casting from sensor origin to endpoints with a 3D Bresenham-type voxel traversal; endpoints updated as occupied (l_occ = 0.85, p = 0.7) and traversed voxels as free (l_free = -0.4, p = 0.4); an endpoint voxel is never freed within the same sweep update |
| 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 | none |
| 地圖表示 | octree of voxels with log-odds occupancy, clamping thresholds, lossless pruning and multi-resolution queries |
| 先驗資訊 | sensor poses from an external source |
| 可輸出幾何 | occupied/free/unknown voxel map at chosen resolution (map files); not a surface model |
| 計算需求 | single core of an Intel Core i7-2600 (3.4 GHz) desktop CPU for all runtime tests; memory accounted for a 32-bit architecture (inner node 40 B, leaf 8 B) |
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 | SICK LMS | 資料集感測器 | FR-079 corridor | 2D laser range finder on a pan-tilt unit; beam range limited to 10 m for this dataset; 66 3D scans, 6 million end points | (Hornung et al., 2013, Sec. 5.2) |
| LiDAR | two fixed sweeping laser scanners (model not reported) | 資料集感測器 | New College (Epoch C) | sweeping to the left and right side of the robot; 14 million end points | (Hornung et al., 2013, Sec. 5.2) |
| LiDAR | laser scanner (model and scanning mechanism not reported) | 資料集感測器 | Freiburg campus | 81 dense 3D scans, 20 million end points, full laser range up to 50 m | (Hornung et al., 2013, Sec. 5.2, Sec. 5.5.1) |
| RGB-D 相機 | Microsoft Kinect | 資料集感測器 | TUM RGB-D freiburg1_360 | hand-held; colored point clouds, 210 million end points | (Hornung et al., 2013, Sec. 3.5.1, Sec. 5.2) |
| 載具平台 | Pioneer2 AT | 資料集感測器 | FR-079 corridor | mobile robot carrying the pan-tilt laser | (Hornung et al., 2013, Sec. 5.2) |
| 運算硬體 | Intel Core i7-2600 | 執行運算平台 | 未標示 | 3.4 GHz; single core used | (Hornung et al., 2013, Sec. 5.5) |
作者報告的優勢與限制
優勢
- ["Explicitly represents free and unknown space, which point clouds and 2.5D maps do not (Sec. 1-2).", "Compact maps via pruning
- with 16 depth levels at 1 cm resolution a map spans 655.36 m per dimension (Sec. 4).", "97.27% to 98.79% of cells agree with the scans, and 95.80% to 98.46% under 80/20 cross-validation (Sec. 5.3, Table 1).", "Freiburg campus at 10 cm: 5162.90 MB for a full 3D grid versus 990.66 MB for the pruned octree and a 13.82 MB lossy map file (Table 2).", "Scans of about 90,000 to 250,000 points integrate in under a second
- the full-resolution Freiburg campus map (1,087,014 occupied and 3,377,882 free leaves) is traversed in 51 ms (Sec. 5.5).", "Pruning with clamping improves compression by up to 44% (Sec. 3.4)."]
限制
- ["Sensor noise and discretization errors produce differing probabilities that interfere with compression relying on identical node values (Sec. 3.4).", "Ray-casting discretization during shallow-angle sensor sweeps can mark occupied cells as free and create holes
- the authors mitigate this by treating a sweep as one point cloud and never freeing an endpoint voxel in the same update (Sec. 5.1, Figs. 10-11).", "Clamping makes compression lossy near probabilities 0 and 1 (Sec. 3.2, Sec. 3.4).", "Remaining map error is attributed to sensor noise, discretization effects or imperfect scan alignment (Sec. 5.3).", "In confined spaces an uncompressed octree can need more memory than an optimally aligned 3D grid, e.g. FR-079 at 10 cm: 10.87 MB versus 10.01 MB (Sec. 5.4, Table 2).", "(inference) The accuracy metric measures agreement with the same or held-out scans, not with an independent survey reference, so it does not bound geometric error."]
營建工程相關證據
未在營建場域驗證；測試場景為弗萊堡大學 079 館走廊（43.7 m x 18.2 m x 3.3 m）、校園戶外（292 m x 167 m x 28 m）與 New College。其「未知空間」表示可用於掃描覆蓋與缺漏判斷，但 Table 1 的準確率只衡量體素佔據狀態與掃描本身的一致性，並非對獨立量測基準的幾何誤差，佔據體素也不是量測級表面（推論）。
原文驗證環境：公開基準、模擬、受控實驗、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 18 個比較組，合計 170 筆紀錄。以下列出本方法紀錄最多的 4 組，其餘 14 組列在最後，並連到性能比較頁。
Hornung et al., 2013 · Table 2 本方法 50 筆
表格設定（擷取紀錄原文）：Memory on a 32-bit architecture: full 3D grid (minimal bounding box, one float per cell) versus OctoMap without compression, pruned, and maximum-likelihood compressed; file sizes for the full probabilistic and the lossy maximum-likelihood binary format; (*) freiburg1_360 voxels include full color (Hornung et al., 2013, Table 2)
Memory w. octree compression [MB], Max. likelih.，FR-079 corridor · 43.7 x 18.2 x 3.3 m, resolution 5 cm
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 Hornung et al., 2013 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 |
|---|---|---|
| OctoMap, maximum-likelihood compression本方法原文提出硬體：32-bit memory accounting | 24.72 MB | (Hornung et al., 2013, Table 2) |
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。
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 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。
資料來源作者報告值(Lim et al., 2021, Table II)
| 方法（原文寫法） | 報告值 | 出處 |
|---|---|---|
| OctoMap - 0.05本方法 | 76.731% | (Lim et al., 2021, Table II) |
| OctoMap - 0.2本方法 | 34.568% | (Lim et al., 2021, Table II) |
| Peopleremover | 37.523% | (Lim et al., 2021, Table II) |
| Removert - RM3 | 85.502% | (Lim et al., 2021, Table II) |
| Removert - RM3+RV1 | 86.829% | (Lim et al., 2021, Table II) |
| ERASOR (Ours)原文提出 | 93.98% | (Lim et al., 2021, 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。
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 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。
資料來源作者報告值(Wang et al., 2025a, Table I)
| 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)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本方法硬體：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) |
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。
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 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。
資料來源作者報告值(Duberg et al., 2024, Table I)
| 方法（原文寫法） | 報告值 | 出處 |
|---|---|---|
| Removert [8] | 99.44% | (Duberg et al., 2024, Table I) |
| ERASOR [9] | 66.7% | (Duberg et al., 2024, Table I) |
| OctoMap [16]本方法 | 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) |
其他比較組
列出其餘 14 個比較組
- Zhang et al., 2023a · Table I
- Duberg et al., 2024 · Table III
- Hornung et al., 2013 · Table 1
- Vizzo et al., 2022 · Table 6
- Vespa et al., 2018 · Table III
- Vespa et al., 2018 · Table IV
- Duberg et al., 2024 · Table II
- Zhang et al., 2023a · Table II
- Vespa et al., 2018 · Text Sec.V-B
- Vizzo et al., 2022 · Table 2
- Vizzo et al., 2022 · Table 4
- Vizzo et al., 2022 · Table 5
- Lim et al., 2021 · Table III
- Hornung et al., 2013 · Text Sec. 5.5.2
來源
Hornung et al., 2013
（2013）OctoMap: an efficient probabilistic 3D mapping framework based on octreesAutonomous Robots, 34(3):189-206
DOI 10.1007/s10514-012-9321-0程式碼
同儕審查已出版已讀全文經典查證後修正
程式碼：https://github.com/OctoMap/octomap（授權：New BSD (octomap library); GPL (octovis viewer), per repository README）。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。