VICP
一般 ICP 假設同一次掃描的點同時量測,但測距儀是逐點依序量測,快速運動時會產生掃描畸變並累積追蹤誤差。VICP 假設掃描期間速度固定,以前後兩次掃描的相對轉換估計測距儀的本體速度,據以校正每個點後重跑 ICP,再更新速度直到收斂;採用以最後一點為基準的反向補償,使位姿不因掃描時間而延遲,並依感測範圍剔除沒有對應的點。作者以 2D Hokuyo 測距儀在模擬與辦公室推車實驗中驗證(Sec. IV 至 VI)。
本頁內容
Estimates rangefinder velocity during ICP iterations to compensate scan distortion caused by sequential point acquisition.
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 rangefinder (Hokuyo URG-04LX, written HOKUYO URG04-LX) |
|---|---|
| 原文測試平台 | simulation、cart pushed by a person carrying the rangefinder and a laptop in an office (7.2 m x 7.8 m) |
| 狀態估計 | ICP with velocity update |
| 資料關聯 | 2D point-to-point closest-point ICP solved by SVD (Algorithm 1), with points outside the pie-shaped sensing area of the previous scan rejected as outliers |
| 時間表示 | constant velocity within a scan |
| 去畸變 | each point transformed with the constant body velocity estimated from the relative ICP transform (backward difference over the scan interval) and re-estimated in an outer loop until the velocity converges; backward compensation uses the last point as reference so the pose is not delayed by the scan time |
| 迴圈閉合 | none |
| 全域最佳化 | none |
| 地圖表示 | 不適用 |
| 先驗資訊 | none |
| 可輸出幾何 | 2D rangefinder trajectory and motion-compensated scans |
| 計算需求 | C++ on a laptop with an Intel dual core 2.0 GHz CPU; total cost grows by less than twice that of ICP because ICP iterations drop as the velocity converges; 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 | HOKUYO URG04-LX歸入:Hokuyo URG-04LX | 方法輸入 | 未標示 | 2D laser scanner; 785 nm; accuracy plus or minus 10 mm; resolution 1 mm; scan angle 240 deg; range 4000 mm; angular resolution 0.36 deg; 100 ms per scan; 50 x 50 x 70 mm; the paper spells the model URG04-LX (the vendor spelling is URG-04LX) | (Hong et al., 2010, Table I, Sec. V, VI) |
| 運算硬體 | laptop with Intel dual core 2.0 GHz CPU | 執行運算平台 | 未標示 | C++ implementation; no timings reported | (Hong et al., 2010, Sec. VI) |
作者報告的優勢與限制
優勢
- Lower return-to-start drift than original ICP in all four office experiments, e.g. 177 mm and 7.28 deg versus 2191 mm and 58.14 deg in experiment 1 (Table II)
- Advantage is largest for faster motion (experiment 2 at about 2.7 m/s average: 65 mm versus 2014 mm) (Sec. VI, Table II)
- Total computation grows by less than twice that of ICP (Sec. IV-C, Fig. 5)
- Backward compensation yields an up-to-date pose (Sec. IV-B)
限制
- Assumes constant velocity during one scan (Sec. IV-B)
- Convergence of the velocity iteration is observed but not proven (Sec. IV-B)
- Real-world accuracy judged only by return-to-start drift because ground truth was unavailable (Sec. VI)
- Tested only with a 2D rangefinder, in simulation and in one small office (Sec. V, VI)
營建工程相關證據
原文未報告
原文驗證環境:模擬、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 1 個比較組,合計 8 筆紀錄。
Hong et al., 2010 · Table II 本方法 8 筆
表格設定(擷取紀錄原文,這些數值分屬表中不同部分):(Hong et al., 2010, Table II)
- Office experiments with cart-carried URG-04LX; drift error on return to the starting point (no ground truth)
- Office experiments; return-to-start drift
R. Error,own office experiments · Experiment 1 (curved path, 16.9 s, about 1.2 m/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 Hong et al., 2010 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(Hong et al., 2010, Table II)
| 方法(原文寫法) | 報告值 | 出處 |
|---|---|---|
| Original ICP | 58.14 deg | (Hong et al., 2010, Table II) |
| VICP本方法原文提出 | 7.28 deg | (Hong et al., 2010, Table II) |
來源
Hong et al., 2010
(2010)VICP: Velocity updating iterative closest point algorithm2010 IEEE International Conference on Robotics and Automation, pp. 1893-1898
DOI 10.1109/robot.2010.5509312
同儕審查已出版已讀全文經典