Lidar-IMU calibration with upsampled preintegration
作者指出 LiDAR 點是逐點取樣而非快照,平台快速移動時會產生運動畸變。本法以高斯過程迴歸對 IMU 讀數上取樣,對每個 LiDAR 點計算預積分量以精確去畸變,並以點到平面距離與 IMU 預積分因子聯合估計外參、IMU 狀態與時間偏移。模擬顯示未使用上取樣預積分時,在快速運動下精度明顯下降。
本頁內容
Gaussian-process upsampled IMU preintegration per LiDAR point enables jointly calibrating LiDAR-IMU extrinsics and time shift while correcting motion distortion.
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 (VLP-16)、IMU (Xsens MTi-3) |
|---|---|
| 原文測試平台 | handheld (moved in front of a room corner)、simulation |
| 狀態估計 | on-manifold factor-graph optimization of IMU poses, velocities, biases, calibration and time shift |
| 資料關聯 | RANSAC (MLESAC) plane detection in every scan; planes tracked between consecutive scans by nearest-neighbour matching of normals after subtracting their centroid; each point assigned to one map plane from the first scan (assumed static) and used in a point-to-plane residual |
| 時間表示 | continuous-time IMU signal via Gaussian-process regression, per-point preintegration |
| 去畸變 | each LiDAR point reprojected with upsampled preintegrated IMU measurements |
| 迴圈閉合 | none |
| 全域最佳化 | batch |
| 地圖表示 | set of planes |
| 先驗資訊 | none |
| 可輸出幾何 | extrinsic calibration and time shift |
| 計算需求 | offline batch in Matlab (Manopt trust-region solver on manifolds; gpml toolbox for GP regression); points capped per plane per scan to shorten processing; runtime not reported; a C++ implementation is planned |
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 | simulated as 16 channels (plus or minus 15 deg), 10 Hz, 240k points per second; FOV reduced to plus or minus 45 deg for the handheld rig; read with the snark driver | (Le Gentil et al., 2018, Sec. III-C, IV-A, IV-B, Fig. 8) |
| 慣性量測單元(IMU) | Xsens MTi3歸入:Xsens MTi-3 | 方法輸入 | 未標示 | low-cost IMU; simulated at 100 Hz; read with the ROS Xsens driver | (Le Gentil et al., 2018, Sec. IV-A, IV-B, V, Fig. 8) |
| RGB-D 相機 | RGB-D camera (RealSense colour camera; exact model not stated) | 參考或真值量測 | 未標示 | rolling-shutter colour camera; not used by the proposed method, only for the chained Kalibr and camera-LiDAR reference calibration | (Le Gentil et al., 2018, Sec. IV-B, Fig. 8) |
作者報告的優勢與限制
優勢
- Upsampled preintegration improves calibration accuracy, especially under fast motion (Sec. IV-A4, Table I)
- Target-less: arbitrary planes detected in the first scan (here a room corner) serve as the calibration target (Sec. I, II-E)
- On a 60 s real sequence (442 of 577 scans used) the result differs from a chained Kalibr IMU-camera plus camera-LiDAR calibration by 3.6 cm and 1.45 degrees despite unsynchronized sensors and partial target views (Sec. IV-B)
- GP regression and high-frequency preintegration reduce the influence of inertial noise on the estimated calibration (Sec. IV-A1)
限制
- Small perturbations of IMU states produced large calibration errors in simulation, so accurate IMU node estimation is needed (Sec. IV-A3)
- Real data from a 60 s sequence in front of a room corner only (Sec. IV-B)
- LiDAR ranging noise is the dominant error source; accuracy drops strongly once LiDAR noise is simulated even with a noise-free IMU (Sec. IV-A1, V)
- The first scan must be static, and the handheld setup reduced the LiDAR FOV to plus or minus 45 degrees (Sec. II-E, III-C)
- No ground truth for the real data, so the comparison with the chained calibration is not conclusive (Sec. IV-B)
營建工程相關證據
未在工地測試;作者建議車輛或無人機可改用倉庫或地下停車場的地板、天花板與牆面作為校正平面(Sec. V),表示方法可利用建築物既有平面,但文中未實測。
原文驗證環境:模擬、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 66 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。
Le Gentil et al., 2020 · Table III 本方法 32 筆
表格設定(擷取紀錄原文):Average computation time over 50 trials for different integration interval lengths; IMU 100 Hz, UPM upsampled to 1 kHz; for UPM and GPM the hyper-parameter training time and the inference time are listed separately (Le Gentil et al., 2020, Table III)
average hyper-parameter training time per preintegrated measurement,simulated IMU trajectories · 1D rotations, interval 0.05 s
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 Le Gentil et al., 2020 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(Le Gentil et al., 2020, Table III)
| 方法(原文寫法) | 報告值 | 出處 |
|---|---|---|
| UPM (upsampled preintegration [17])本方法硬體:laptop with Intel i5-6300U CPU at 2.40 GHz and 24 GiB RAM; single-thread Matlab code, not optimised | 294 ms | (Le Gentil et al., 2020, Table III) |
| GPM原文提出硬體:laptop with Intel i5-6300U CPU at 2.40 GHz and 24 GiB RAM; single-thread Matlab code, not optimised | 356 ms | (Le Gentil et al., 2020, Table III) |
Le Gentil et al., 2020 · Table II 本方法 16 筆
表格設定(擷取紀錄原文):Simulated fast trajectories, 100 trials, fixed query rates of 1 to 20 Hz; average absolute error of the preintegrated measurement (the paper calls it absolute pose error); units as printed: mrad and mm for 1D rotations, 'rad' and mm for 3D rotations (Le Gentil et al., 2020, Table II)
average absolute position error of preintegrated measurement,simulated IMU trajectories · 1D rotations, query rate 20 Hz
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 Le Gentil et al., 2020 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(Le Gentil et al., 2020, Table II)
| 方法(原文寫法) | 報告值 | 出處 |
|---|---|---|
| PM (on-manifold preintegration [14]) | 5.61 mm | (Le Gentil et al., 2020, Table II) |
| UPM (upsampled preintegration [17])本方法 | 0.6 mm | (Le Gentil et al., 2020, Table II) |
| GPM原文提出 | 0.02 mm | (Le Gentil et al., 2020, Table II) |
Le Gentil et al., 2020 · Table I 本方法 8 筆
表格設定(擷取紀錄原文):Simulated random sinusoidal trajectories, integration interval 1 to 5 s, 100 trials; average relative error with respect to travelled linear or angular distance; IMU 100 Hz, accelerometer noise sd 0.02 m/s2, gyroscope noise sd 0.002 rad/s, UPM upsampled to 1 kHz (Le Gentil et al., 2020, Table I)
average relative rotation error of preintegrated measurement,simulated IMU trajectories · 1D rotations, slow motion
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 Le Gentil et al., 2020 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(Le Gentil et al., 2020, Table I)
| 方法(原文寫法) | 報告值 | 出處 |
|---|---|---|
| PM (on-manifold preintegration [14]) | 0.253% | (Le Gentil et al., 2020, Table I) |
| UPM (upsampled preintegration [17])本方法 | 0.038% | (Le Gentil et al., 2020, Table I) |
| GPM原文提出 | 0.028% | (Le Gentil et al., 2020, Table I) |
Le Gentil et al., 2018 · Table I 本方法 4 筆
表格設定(擷取紀錄原文,這些數值分屬表中不同部分):(Le Gentil et al., 2018, Table I)
- Simulation with exact known IMU poses and noise-free measurements; mean calibration error over 10 Monte Carlo runs; normal motion 0.2 to 0.53 Hz sines, fast motion 1.53 Hz
- Simulation with exact IMU poses and noise-free data
translational calibration error ep,simulation · normal motion
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 Le Gentil 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。
資料來源作者報告值(Le Gentil et al., 2018, Table I)
| 方法(原文寫法) | 報告值 | 出處 |
|---|---|---|
| (i) with upsampled preintegrated measurements本方法原文提出 | 0.00057 m | (Le Gentil et al., 2018, Table I) |
| (ii) without upsampled preintegrated measurements | 0.01 m | (Le Gentil et al., 2018, Table I) |
其他比較組
來源
Le Gentil et al., 2018
(2018)3D Lidar-IMU Calibration Based on Upsampled Preintegrated Measurements for Motion Distortion Correction2018 IEEE International Conference on Robotics and Automation (ICRA), pp. 2149-2155
同儕審查已出版已讀全文近十年
相關版本
- 預印本:author accepted manuscript (UTS OPUS) https://opus.lib.uts.edu.au/bitstream/10453/122728/4/OCC-113281_AM.pdf