- 學位論文
提升數位影像相關法的量測精度並應用於車輛追蹤與機械手臂的三維量測
Improving the Accuracy of Digital Image Correlation and Applying it to Vehicle Tracking and 3D Measurements of Robotic Arms
摘要
數位影像相關法（digital image correlation, DIC）是一種非接觸式的光學測量技術，可應用於跨尺度與跨領域的工程問題上。其原理為透過數位影像序列分析，追蹤目標表面的灰階特徵，取得位移場、速度、加速度、熱伸長、熱應變等物理量。二維DIC系統使用單相機進行面內量測；三維DIC系統使用雙相機量測三維空間之全場資訊。本文加強了本實驗室自行開發的DIC量測精度與計算速度，提升DIC技術於各項產業實際問題之泛用性，並將其應用於兩項工業界所面臨之議題：複雜交通場景之車輛追蹤與 ISO 9283 標準化機械手臂三維運動性能量測。 在車輛追蹤上，DIC在過往遇到兩個困難點：（1）複雜街景之背景干擾，例如柏油路上所印的交通標誌（2）追蹤目標形貌灰階值變化過大，例如車輛駛近、駛遠與轉彎造成特徵放大縮小或旋轉。以上兩點皆會導致過往DIC系統無法正確追蹤標的物而導致追丟目標。本文提出扣除「平均影像」的手法，去除背景對DIC特徵搜索過程之影響；以及運用「更新模板法」，有效克服特徵劇烈變化而產生的追蹤失效問題。 在三維 DIC的測量能力上，本文提出一種適合DIC三維追蹤的特殊標記，大幅提升影像深度方向的測量性能，避免因深度方向運動時特徵放大縮小變化所導致的追蹤失敗，大幅增加三維DIC技術的量測範圍。將DIC量測技術與三次元量床進行比較後，確立DIC沿影像高度、寬度和深度方向的誤差分別落在0.2%、0.45%和2%。本文也針對三維 DIC量測技術之系統架設和校正鏡頭畸變提出相關校正流程與規範。 在機械手臂的性能量測上，DIC同時滿足手臂大範圍與高精密量測的需求，本文遵循ISO 9283機械手臂性能量測規範，將三維 DIC量測結果與業界常用的雷射追蹤儀比對：絕對精度差異小於0.064 mm，重複性精度差異小於0.01 mm。此結果顯示相較於昂貴的雷射追蹤儀，DIC技術同樣可提供大範圍運動量測，甚至在重複精度上呈現更好的穩健性。
並列摘要
Digital Image Correlation (DIC) is a non-contact measuring technique for diverse engineering problems. By analyzing image series and tracking a region of interest (ROI) target, DIC can derive physical quantities such as displacement and velocity. In this thesis, the accuracy of our self-developed DIC system (in MATLAB) is improved; then it is applied to two different problems: vehicle tracking in complex traffic scenarios and 3D standardized robotic arm performance evaluation according to ISO 9283. In vehicle tracking, two obstacles were identified and overcame in this thesis: (1) unwanted interference in image background (e.g. traffic markings on pavement) and (2) ROI features, cars in this case, change drastically as cars turn or drive near/away. Background interference is eliminated by utilizing an “average image”. ROI change is overcome by applying the “updating template method” proposed in this thesis. In terms of 3D DIC, measuring capability is improved greatly, especially along depth direction measurements. ROI change for motions in depth direction is overcome by using a special tracking marker (2×2 checkerboard). By comparing with a coordinate measuring machine, measuring error is found to be 0.2, 0.45 and 2% along height, width and depth direction of DIC camera system. This thesis also provides standardized guidelines for camera setup and lens calibration processes. In robotic arm performance evaluation, DIC results match well with laser tracker, a prominently used but expensive machine for robotic arm calibration. Absolute precision differences were less than 0.064 mm, and repeatability precision differences were less than 0.01 mm. Between experiments, DIC outperforms laser tracker in repeatability.