- 學位論文
Feature-based 顯微環境下的物件辨識與3D追蹤
Feature-based Object Recognition and Three-Dimensional Tracking in Microscopic Environment
摘要
本論文開發一項能夠運用於光學顯微鏡下的即時計算微小物件六自由度位移量的影像處理技術。 本論文以Scale Invariant Feature Transform (SIFT)技術為基礎，鎖定目標物影像上，相同特徵點在連續影像的變化，可推算目標物精確的位移、旋轉、Scale資訊，所謂的affine transform。本論文創新之處在於:即使在影像模糊的狀況下，仍可根據scale的變化，精確地求出目標物的距離量。經過實驗證明:在20個景深範圍內，物距與特徵的Scale之間存在極佳的線性關係，可從Scale的變化量準確的計算出目標物的物距。利用此項成果，作者開發一套即時顯微影像追蹤系統，能夠察覺微小目標物六個自由度位移，透過奈米壓電驅動顯微鏡物鏡控制物距，維持目標物在固定的影像位置及對焦距離。 本論文所提出的方法成功地應用於景深4.4 micrometer的10倍光學顯微鏡下追蹤微小物件的六個自由度運動，定位精度高於0.5 micrometer。
並列摘要
This paper presents a method to calculate the motion of the object in six degrees of freedom with microscope in real time. This paper is based on “Scale Invariant Feature Transform”. It is called “affine transform” that can calculate the translation, rotation and scale of the target according to the change of the same features in continuous images. The innovation of this paper is that we can accurately calculate the depth of the target according to the change of scale even in the blurred image. Proved through the experiment, within the range of 20 times of depth of field, there is an excellent linear relation between depth and scale. It could let us calculate the depth of the target accurately by the variation of scale. With this achievement, the author develops a real time image tracking system. This system can calculate the motion of the micro-component with six degrees of freedom and keep the micro-component at the fixed image position and focus plane by piezo-driven positioner. The proposed method is successfully applied to track the six DOF motion of a miniature component using a 10X objective lens and 4.4 micrometer depth of field. The steady-state error is less than 0.5 micrometer.