利用旋轉式單眼相機之物件定位系統設計與實現

Abstract

本篇論文利用旋轉式攝影機設計出一個新的物件定位系統，其可估測之距離為一至兩公尺，且誤差值約在10%以內。此系統由三個部分組成，分別是伺服馬達MX-106r、羅技公司的視訊攝影機C920以及一台電腦。在開始執行物件定位的演算法前，必須先進行兩項實驗來辨識與校準系統的參數。第一個實驗是自動根據攝影機的廣角度數(FOV)，藉由計算物體位於每一個角度會顯示在影像中的位置，求得一函數用以表示物體位置相對於攝影機的水平視角與仰角。第二個實驗則是在校準攝影機的旋轉半徑，在本篇論文中設計一個白線搭配黑底的圖片作為校準目標物，應用Canny邊緣檢測和霍夫找線演算法，以半自動的方式對攝影機前方的校準目標物完成校準。由於此系統只使用一個攝影機與雙眼視覺定位不同，並不存在雙眼間的長度(LOS)，因此利用旋轉攝影機的方式，以拍攝影像的兩點位置間距離作為該常數。最後根據上述的結果以及利用物體於兩張影像中位置所求得之視角，以正弦與餘弦定理進行計算，即可估測物體在3-D空間中的位置，並以圓柱坐標系表示其結果P=(R, alpha, Z)。

This thesis designs and implements a novel object positioning system using a rotary monocular camera and it can estimate the distance ranging from 100 cm to 200 cm within ten percentage error rate. The system proposed in this thesis consists of three components, MX-106r servo motor, Logitech C920 web camera and a computer. It needs two experiments before estimating the position of the object. Based on the field of view (FOV) of the camera, the first experiment can automatically obtain the object’s angle in terms of the number of pixels of the reference point deviated from the central line degree-by-degree. The second one is the calibration of the rotation radius. With the application of the canny edge detection and the Hough line detection, the system can semi-automatically completes the calibration of the rotation radius. Due to the use of one camera, there is no length of sight (LOS) exists in the system naturally. Therefore, the system creates the LOS by rotating the camera, which is the distance between two positions where the system took images. Based on the previous results, the system can obtain the object’s angles which are converted from positions in two images and use the law of sine and cosine to estimate the object’s position P=(R, alpha, Z) in cylinder coordinate.