Incidence-angle LiDAR bias model
作者質疑 LiDAR 量測為零均值高斯雜訊的常見假設，指出與入射角及距離相關的偏差會造成可預期的定位漂移，例如直線隧道的地圖會依靠近哪一側牆而彎曲。本文以回波波形建模解釋此偏差，於實驗裝置量測三款 LiDAR 的偏差，發現高入射角下可達 20 cm，並用模型修正量測以改善地圖與漂移。
本頁內容
Models incidence-angle and range dependent LiDAR bias via return-waveform physics, measures it for three LiDARs (up to 20 cm), and corrects maps and drift.
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 LiDAR (SICK LMS151)、3D LiDAR (Velodyne HDL-32E, Robosense RS-LiDAR-16) |
|---|---|
| 原文測試平台 | mobile robot carrying the LiDAR in a university tunnel (robot type not stated in the text read)、controlled test bench (linear motion table) |
| 狀態估計 | 不適用 (measurement model and correction) |
| 資料關聯 | 不適用 |
| 時間表示 | 不適用 |
| 去畸變 | 不適用 |
| 迴圈閉合 | 不適用 |
| 全域最佳化 | none |
| 地圖表示 | 不適用 |
| 先驗資訊 | none |
| 可輸出幾何 | bias-corrected range measurements and maps |
| 計算需求 | closed-form correction implemented in libpointmatcher; authors state the computation is lightweight and does not affect mapping performance; no timings reported |
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 LMS151歸入：SICK LMS-151 | 方法輸入 | 未標示 | 2D LiDAR; fitted aperture half-angle 0.43 deg (Table I) | (Laconte et al., 2019, Sec. IV, V, Table I, Fig. 1) |
| LiDAR | Velodyne HDL-32E | 方法輸入 | 未標示 | 3D LiDAR; fitted aperture half-angle 0.085 deg (Table I); mounted on the robot for the 3D tunnel map | (Laconte et al., 2019, Sec. IV, V, Table I, Fig. 9) |
| LiDAR | Robosense RS-LiDAR-16 | 方法輸入 | 未標示 | 3D LiDAR lent by Robosense; fitted aperture half-angle 0.085 deg (Table I) | (Laconte et al., 2019, Sec. IV, V, Table I) |
| 其他 | Renishaw XL-80 interferometer | 參考或真值量測 | 未標示 | ground-truth board distance with accuracy of the order of 1 micrometre | (Laconte et al., 2019, Sec. IV, Figs. 5-6) |
| 其他 | linear motion table (rail) with rotating wooden board | 參考或真值量測 | 未標示 | rail of about 30 m; board 0.6 x 0.6 m on a graduated disc; 12 board orientations (0 to 60 deg in 10 deg steps, 65 to 85 deg in 5 deg steps) at 8 depths from 1 to 10 m; 45 s per setting; metrology room at 20 C | (Laconte et al., 2019, Sec. IV, Fig. 5) |
作者報告的優勢與限制
優勢
- Bias correction reduced map bending and localization drift in a long straight underground corridor example (Sec. I, Fig. 1)
- A physical waveform model with only two per-sensor scale factors was fitted to bench data at depths 1 to 10 m and incidence angles 0 to 85 degrees for three LiDARs (Sec. IV, V, Table I)
- Unlike the empirical model of Pfister et al., the analytic model captures the angle-dependent slope of bias with depth, and it needs fewer samples and extrapolates better than a non-parametric kernel fit (Sec. V)
限制
- Normals used for incidence angle should be estimated iteratively with bias taken into account (Sec. VI, future work)
- Map improvements are shown only qualitatively; no quantitative drift or map error is reported (Sec. V, Figs. 1 and 9)
- 3D corrections were noisy at long range because normals from sparse HDL-32E data give unreliable incidence angles (Sec. V, Fig. 9)
- Bench measurements beyond 10 m were unreliable because the beam exceeded the 0.6 m board at high angles, and residual setup misalignment may remain (Sec. IV)
營建工程相關證據
未在工地測試；隧道與長走廊的高入射角量測與營建地下工程及室內走廊高度相關（推論）。
原文驗證環境：受控實驗、地下或隧道、獨立參考量測
報告的性能數據
性能數據仍在分批查證，目前尚未收錄此方法的報告值。
來源
Laconte et al., 2019
（2019）Lidar Measurement Bias Estimation via Return Waveform Modelling in a Context of 3D Mapping2019 International Conference on Robotics and Automation (ICRA), pp. 8100-8106
DOI 10.1109/icra.2019.8793671arXiv 1810.01619
同儕審查已出版已讀全文近十年查證後修正
相關版本
- 預印本：Lidar Measurement Bias Estimation via Return Waveform Modelling in a Context of 3D Mapping https://arxiv.org/abs/1810.01619