Passive synchronization
許多商用感測器不支援同步，只能在資料抵達主機時打時間戳記，而緩衝與非即時作業系統造成的抖動在高負載時可達數百毫秒。本文利用延遲不可能為負的因果關係，搭配感測器時鐘速率漂移的上界模型，以取最大值的規則估計感測器與主機時鐘的偏移，再還原每筆資料的主機時間。演算法可單向即時執行或前後兩次處理，作者證明結果不會劣於直接以抵達時間打戳記，但實驗僅為合成資料。
本頁內容
Passive algorithm reducing timestamp error for sensors lacking synchronization support by modelling host-induced jitter.
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 | generic sensors without synchronization support (examples in text: SICK and Hokuyo LIDARs and Xsens IMUs behind USB-to-serial converters)、deployment context only: 12 SICK LIDARs, a Velodyne HDL-64E, 15 Delphi ACC radars and an Applanix IMU/GPS on MIT's DARPA Urban Challenge vehicle (not evaluated in this paper) |
|---|---|
| 原文測試平台 | simulation (synthetic timing data only) |
| 狀態估計 | max-rule lower-bound estimate of the sensor-to-host clock offset using causality (latency is non-negative) and a bounded clock-rate drift model with parameters alpha1 and alpha2; two-pass O(N) algorithm (causal forward pass plus optional backward pass) |
| 資料關聯 | 不適用 |
| 時間表示 | per-message recovery of host time t = p - A(p), with a time-varying clock offset A(p) bounded by the rate-drift model |
| 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 | user-specified bounds on the sensor-to-host clock-rate error (alpha1, alpha2) that define the drift model; optionally a known minimum system latency; sensor timestamps or a regular message rate acting as a clock |
| 可輸出幾何 | corrected timestamps |
| 計算需求 | O(1) per observation in causal online mode and O(N) for the bidirectional two-pass version; a few lines of code; no real-time OS required |
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 HDL-64E | 方法輸入 | 未標示 | one unit on MIT's DARPA Urban Challenge vehicle, synchronized with this algorithm; deployment context, not evaluated in the paper | (Olson, 2010, Sec. I) |
| LiDAR | SICK LIDAR (12 units; model not stated) | 方法輸入 | 未標示 | deployment context on the DARPA Urban Challenge vehicle; not evaluated | (Olson, 2010, Sec. I) |
| GNSS 接收器 | Applanix IMU/GPS | 方法輸入 | 未標示 | deployment context on the DARPA Urban Challenge vehicle; not evaluated | (Olson, 2010, Sec. I) |
| 雷達 | Delphi ACC Radar (15 units) | 方法輸入 | 未標示 | deployment context on the DARPA Urban Challenge vehicle; not evaluated | (Olson, 2010, Sec. I) |
| 載具平台 | MIT DARPA Urban Challenge vehicle | 方法輸入 | 未標示 | vehicle on which the algorithm was originally developed and used | (Olson, 2010, Sec. I) |
作者報告的優勢與限制
優勢
- Provably no worse than naive arrival-time stamping; no real-time OS needed (abstract)
- Never places an observation earlier than it occurred and never does worse than naive stamping (Claims 1 and 2, Sec. III-C)
- In synthetic tests with uniform latency up to 0.5 s at 1 s intervals, naive stamping averaged 0.25 s error while both variants were substantially lower, the bidirectional one best (Fig. 6 caption, Sec. IV)
限制
- Validated only on synthetic data (reviewer observation, Sec. IV)
- Performance approaches the no-synchronization case as sensor clock drift grows (Fig. 6 caption, Sec. IV)
- Offset error increases as the time between observations grows (Fig. 7, Sec. IV)
- The causal online variant is less accurate than the non-causal bidirectional variant (Sec. III-D, IV)
- Needs sensor timestamps or a regular message rate acting as a clock, and drift bounds must be specified (Sec. III)
營建工程相關證據
未在工地或測繪載具上評估。文中舉例：以每秒 90 度旋轉、觀測 10 m 外物體時，10 ms 的同步誤差即造成 15.7 cm 投影誤差；經 USB 轉序列埠連接的 LiDAR 與 IMU 也會引入可變延遲。這對以低成本感測器自組的工地掃描設備具參考價值（推論）。
原文驗證環境：模擬
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
Olson, 2010
（2010）A passive solution to the sensor synchronization problem2010 IEEE/RSJ International Conference on Intelligent Robots and Systems, pp. 1059-1064
同儕審查已出版已讀全文經典