SLAM 點雲圖鑑Point clouds for construction
M-01 · 2018

StaticFusion

StaticFusion 是針對動態環境的 RGB-D 稠密 SLAM，同時估計相機運動與影像中哪些區域靜止。每張影像先以 K-means 依三維座標分成幾何群集，再與由靜態面元地圖渲染出的預測影像做光度與幾何直接對齊；每個群集有一個 0 至 1 的靜態分數，用來加權其殘差，分數則由殘差門檻、相鄰群集平滑與深度差先驗共同決定，並以 IRLS 交替求解位姿與分數。融合時只寫入靜態部分，每個面元以對數勝算累積被靜態點重複觀測的可信度，持續被動態點匹配的面元會被移除，因此地圖只保留背景結構，並把這個背景作為下一影格分割所需的時間資訊。

里程計＋局部建圖近十年已讀全文同儕審查已出版視覺、視覺慣性與 RGB-D 稠密 SLAM

閱讀版本
IEEE Xplore HTML full text of the ICRA 2018 version of record with Tables I and II viewed as publisher images
全文取得途徑
國立臺灣大學機構訂閱
本頁內容

Dense RGB-D SLAM for dynamic scenes that jointly estimates camera motion and per-cluster static scores by score-weighted photometric and geometric alignment against a rendered static surfel map, and fuses only static data with per-surfel viability so the map keeps only the background.

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RGB-D camera, registered RGB-D images at QVGA 320x240 (TUM Freiburg sequences and two hand-held recordings; hand-held camera model not named)
原文測試平台hand-held RGB-D camera (two authors' sequences, including a 'selfie' sequence, Sec. VII-B)、TUM Freiburg benchmark sequences (capture platform not described in the paper)
狀態估計Joint minimization over the camera twist and per-cluster static scores b in [0, 1]: Cauchy-robust photometric and geometric residuals between the current RGB-D frame and a prediction rendered from the static surfel map, weighted by b, plus a residual-threshold term, spatial regularization between contiguous clusters and a depth-difference prior; IRLS for the twist with closed-form b after each iteration, coarse-to-fine
資料關聯Dense direct alignment (warping current pixels into the rendered model prediction); scene split into K geometric clusters by K-means on 3D coordinates; per-pixel segmentation derived from cluster scores
時間表示discrete poses
去畸變不適用 (RGB-D input)
迴圈閉合none described
全域最佳化none described
地圖表示surfel map (Keller et al. model through the ElasticFusion implementation) in which each surfel carries a viability value accumulated as log-odds of matches with static input points; surfels with viability below 0.5 for more than 10 consecutive frames are removed and free-space violations are cleaned
先驗資訊none
可輸出幾何static-background surfel map (coloured for visualization), camera trajectory and per-frame static/dynamic segmentation
計算需求Intel Core i7-3770 at 3.40 GHz with GeForce GTX 1070 GPU; about 30 ms per frame at QVGA (Sec. VII, VIII)

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
RGB-D 相機hand-held RGB-D camera (model not reported)方法輸入未標示two recorded sequences: person interacting with objects, and a selfie sequence with the camera pointing at its carrier(Scona et al., 2018, Sec. VII-B)
運算硬體workstation with Intel(R) Core(TM) i7-3770 CPU at 3.40GHz and GeForce GTX 1070 GPU執行運算平台未標示Ubuntu 16.04; all compared methods run on it(Scona et al., 2018, Sec. VII)

作者報告的優勢與限制

優勢

  • Lowest ATE on the highly dynamic TUM walking sequences, for example 1.4 cm on fr3/walk_static and 12.7 cm on fr3/walk_xyz versus 29.3 cm and 90.6 cm for ElasticFusion (Table II)
  • Accuracy comparable to ElasticFusion in static sequences (Tables I and II; Sec. VII-A)
  • In a 9.5 m hand-held 'selfie' sequence the final drift was 1.5 cm versus 1.03 m for ElasticFusion and 0.88 m for Co-Fusion (Sec. VII-B)
  • About 30 ms per frame (Sec. VIII)

限制

  • Initialization needs the first 1 to 2 s to contain no more than about 20 to 30% moving elements; the method has high error (39.1 cm ATE) on fr3/walk_halfsphere, whose start is highly dynamic (Sec. VI-A, VII-A, VIII)
  • Struggles if significant portions of the estimated static scene begin to move (Sec. VIII)
  • Static/dynamic segmentation is solved per cluster rather than per pixel, an approximation (Sec. III)
  • Less accurate than ElasticFusion on some static sequences, for example fr1/plant ATE 11.3 cm versus 5.3 cm (Table II)
  • Evaluated with a desktop GPU (GTX 1070) on a GPU-based ElasticFusion code base (Sec. V, VII)

營建工程相關證據

論文未在施工現場測試，評估為 TUM (Freiburg) 室內序列與兩段手持錄製的室內序列（人員與物件移動、自拍式移動）。只把靜態背景融入地圖並持續移除被動態點匹配的面元，對有人員走動的施工中室內掃描有參考價值；但作者指出初始 1 至 2 秒若動態元素超過約 20% 至 30% 便可能失敗，因此工地掃描起始時宜避開人員密集的畫面（推論）。

原文驗證環境：公開基準、受控實驗

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 6 個比較組，合計 65 筆紀錄。以下列出本方法紀錄最多的 4 組，其餘 2 組列在最後，並連到性能比較頁。

Palazzolo et al., 2019 · Table III 本方法 24 筆

指標Absolute Trajectory Error (RMS) [m]

表格設定（擷取紀錄原文）：Bonn RGB-D Dynamic Dataset (24 highly dynamic scenes, ASUS Xtion Pro LIVE, OptiTrack Prime 13 ground truth); ATE RMS; 'o box' = obstructing box, 'no box' = nonobstructing box; all methods run by the authors with default parameters (Palazzolo et al., 2019, Table III)

Absolute Trajectory Error (RMS) [m]，Bonn RGB-D Dynamic Dataset · balloon

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 Palazzolo et al., 2019 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：均方根誤差（RMSE）；對齊方式：原文未報告；單位：m；場景：indoor test room with people manipulating boxes, balloons or crowding the camera

資料來源作者報告值(Palazzolo et al., 2019, Table III)

數值與出處
方法（原文寫法）報告值出處
Ours (ReFusion)原文提出0.175 m(Palazzolo et al., 2019, Table III)
SF (StaticFusion)本方法0.233 m(Palazzolo et al., 2019, Table III)
DS (G) (DynaSLAM geometric)0.05 m(Palazzolo et al., 2019, Table III)
DS (N+G) (DynaSLAM neural network + geometric)0.03 m(Palazzolo et al., 2019, Table III)

Scona et al., 2018 · Table I 本方法 22 筆

表格設定（擷取紀錄原文）：TUM (Freiburg) RGB-D sequences grouped as static (fr1), low dynamic (fr3/sit) and high dynamic (fr3/walk) environments; StaticFusion and VO-SF at QVGA, ElasticFusion and Co-Fusion at their default VGA; fr3/walk_halfsphere* skips the first 5 s of high dynamics; relative pose error per second (Scona et al., 2018, Table I)

Trans. RPE RMSE (cm/s)，TUM RGB-D (Freiburg) · fr1/xyz

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。

  • 未報告（沒有數值，不是 0）

這些是 Scona et al., 2018 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：均方根誤差（RMSE）；對齊方式：原文未報告；單位：cm/s；場景：indoor office scenes, static to highly dynamic (walking people), RGB-D camera (carrying mode not stated in the paper)

資料來源作者報告值(Scona et al., 2018, Table I)

數值與出處
方法（原文寫法）報告值出處
VO-SF (Jaimez et al. joint visual odometry and scene flow)2.1 cm/s(Scona et al., 2018, Table I)
EF (ElasticFusion)1.9 cm/s(Scona et al., 2018, Table I)
CF (Co-Fusion)2.3 cm/s(Scona et al., 2018, Table I)
BaMVO (Kim et al.)無數值未報告註記（擷取紀錄）：not reported (BaMVO shown only for sequences evaluated in its original publication)(Scona et al., 2018, Table I)
SF (StaticFusion)本方法原文提出2.3 cm/s(Scona et al., 2018, Table I)

Scona et al., 2018 · Table II 本方法 11 筆

指標Trans. ATE RMSE (cm)

表格設定（擷取紀錄原文）：TUM (Freiburg) RGB-D sequences grouped as static (fr1), low dynamic (fr3/sit) and high dynamic (fr3/walk) environments; StaticFusion and VO-SF at QVGA, ElasticFusion and Co-Fusion at their default VGA; fr3/walk_halfsphere* skips the first 5 s of high dynamics (Scona et al., 2018, Table II)

Trans. ATE RMSE (cm)，TUM RGB-D (Freiburg) · fr1/xyz

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 Scona et al., 2018 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：均方根誤差（RMSE）；對齊方式：原文未報告；單位：cm；場景：indoor office scenes, static to highly dynamic (walking people), RGB-D camera (carrying mode not stated in the paper)

資料來源作者報告值(Scona et al., 2018, Table II)

數值與出處
方法（原文寫法）報告值出處
VO-SF (Jaimez et al. joint visual odometry and scene flow)5.1 cm(Scona et al., 2018, Table II)
EF (ElasticFusion)1.2 cm(Scona et al., 2018, Table II)
CF (Co-Fusion)1.4 cm(Scona et al., 2018, Table II)
SF (StaticFusion)本方法原文提出1.4 cm(Scona et al., 2018, Table II)

Palazzolo et al., 2019 · Table II 本方法 6 筆

指標Absolute Trajectory Error (RMS) [m]

表格設定（擷取紀錄原文）：TUM RGB-D dynamic scenes; ATE RMS; ReFusion uses virtual depth from 10 frames (about 0.3 s delay) to fill invalid depth; ReFusion, SF and MF are dense, DynaSLAM is feature-based; StaticFusion lost track on walking halfsphere (Palazzolo et al., 2019, Table II)

Absolute Trajectory Error (RMS) [m]，TUM RGB-D · sitting static

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 Palazzolo et al., 2019 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：均方根誤差（RMSE）；對齊方式：原文未報告；單位：m；場景：indoor office with seated or walking people, handheld RGB-D

資料來源作者報告值(Palazzolo et al., 2019, Table II)

數值與出處
方法（原文寫法）報告值出處
Ours (ReFusion)原文提出0.009 m(Palazzolo et al., 2019, Table II)
SF (StaticFusion)本方法0.014 m(Palazzolo et al., 2019, Table II)
DS (G) (DynaSLAM geometric)0.009 m(Palazzolo et al., 2019, Table II)
DS (N+G) (DynaSLAM neural network + geometric)0.007 m(Palazzolo et al., 2019, Table II)
MF (MaskFusion, values from its paper)0.021 m(Palazzolo et al., 2019, Table II)

其他比較組

列出其餘 2 個比較組

來源

  • Scona et al., 2018

    Raluca Scona, Mariano Jaimez, Yvan R. Petillot, Maurice Fallon, Daniel Cremers（2018）StaticFusion: Background Reconstruction for Dense RGB-D SLAM in Dynamic Environments2018 IEEE International Conference on Robotics and Automation (ICRA), pp. 3849-3856

    DOI 10.1109/icra.2018.8460681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/raluca-scona/StaticFusion（授權：GPL-3.0 (GPL-LICENSE.txt), with reused ElasticFusion parts under the ElasticFusion licence (LICENSE.txt checked)）。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