三維影像之重建與追蹤系統之研究

Study toward 3D Image Reconstruction and Tracking System

田鈞豪(Jun-Hao Tian)

指導教授 ： 陳建祥

國立清華大學/工學院/動力機械工程學系/碩士(2007年)

摘要

本研究以個人電腦為基本平台，搭配影像處理方法與立體視覺，以兩台具備水平與垂直旋轉方向的攝影機取像，然後建構出一即時立體影像追蹤系統，來設計實驗。實驗中以兩攝影機同步擷取成對影像，經過影像處理方法分離目標物與背景，藉此框選目標物位置，接著根據極線幾何原理來進行目標物在兩影像中對應特徵點的比對，測量目標物與攝影機光學中心的偏差，並對攝影機進行校正以得知相機參數，之後即可計算目標物的立體影像深度，並即時鎖定目標物進行追蹤。
實驗主要分成兩個部分：第一個部分『即時影像追蹤』，目的在使攝影機能確實的把目標物鎖定在影像範圍內，應用移動邊緣檢測法取得目標物座標及位移方向，送出致動訊號讓攝影機的步進馬達旋轉，將目標物鎖定於影像中心，以達到追蹤效果。第二個部分『3D立體視覺』，目的在取得目標物與攝影機之間的距離，使用各種影像處理方法來簡化影像，測量左右成對影像的偏移距離計算目標影像深度，並探討結果與實際誤差及可能原因。

關鍵字

三維影像重建 ； 影像追蹤 ； 電腦視覺 ； 影像處理

並列摘要

A real-time object tracking system with stereo vision and an image processing method on a PC-based platform was developed. We designed the experimental setup by combining two binocular webcams which can perform pan and tilt, independently. First, the binocular webcams intercept the paired-images simultaneously, then the system separates the target object and the background from the paired-images by an image processing method. Finally, the system locates the target object by using the paired-images for stereo matching. After camera calibration and measuring the difference between the target object and the optical axes, the system can calculate the depth between the target object and the webcams, then commands the webcams to track the object in real-time.
There are two experimental verifications were performed. The first experiment is the reconstruction of a three-dimensional stereo vision from 2-D images. The objective of this part is to calculate the distance between the webcams and the target object. We try to simplify the paired-images by using various image processing methods at the outset. Measuring the difference between the paired-images in order to calculate the depth of the target object and discussing every possible reason that could have caused the error between the result and the correct data in the end. The second experiment is a real-time image tracking system. The objective of this experiment is to command the webcams rotating in the direction of the target object. By applying the moving target shifting method, we measure the geometric center of the target object and let the webcams rotate in the direction of the target object.

並列關鍵字

3D reconstructiion ； feature tracking ； computer vision ； image processing