利用多攝影機進行移動物三維定位及追蹤

Objects Positioning and Tracking Using Multiple Active Video Cameras

賴丙全(Bing-Cyuan Lai)

指導教授 ： 陳繼藩

國立中央大學/工學院/土木工程學系/碩士(2007年)

摘要

利用攝影機進行區域偵測在日常生活中已有相當多應用，本研究利用兩部PTZ攝影機組成的定位追蹤系統，在不需使用數值高程模型的情況下，由攝影機視訊資料中自動化萃取出移動物體在物空間的三維移動軌跡。本研究內容可分為兩階段:第一階段為建立PTZ攝影機像物空間轉換模式，系統中的各PTZ攝影機經由一率定程序，建立像平面與物空間的雙向轉換模式，同時也建立不同攝影機間的空間對應關係;第二階段為移動物體的定位追蹤，使用漸進式背景影像相減法，由攝影機視訊資料中偵測出移動物體在像平面上的位置，再經過匹配和空間前方交會得到移動物體的物空間坐標，最後產生移動物體的三維移動軌跡。實驗部份分別對兩階段成果進行測試，第一階段是以靜態影像及檢核點進行測試，結果顯示能夠正確地進行像平面與物空間之間的雙向轉換，並且能夠交會出目標點在物空間的坐標;第二階段以多種情境的視訊資料進行測試，包含PTZ攝影機固定角度或會改變角度、各攝影機Zoom為相同或不同、移動物體為單個或多個等情況，結果顯示在上述各種情況下，本研究都能由視訊資料中自動化計算出移動物體在物空間的三維坐標，並藉時序關係構成三維的移動軌跡。

關鍵字

PTZ攝影機 ； 三維軌跡 ； 追蹤 ； 定位 ； 移動物體 ； 多攝影機

並列摘要

Object tracking using video camera is an important task for many applications in the field of visual surveillance. In the past, most of studies on object tracking were focused on the issues of finding the trajectories of the moving objects on a two-dimensional (2-D) image plane. However, the trajectories on the image plane are insufficient to represent the complete motions of the moving objects in the object space. In theory, if an accurate digital elevation model (DEM) exists, the collinearity conditions can be used to perform the coordinate transformation between the image plane and the object space. However, it is generally not easy to obtain the accurate DEM for the area under monitor. In this study, multiple video cameras and the concept of epipolar geometry are used to perform the coordinate transformation when the DEM is not available. The main objective of the study is to provide three-dimensional (3-D) trajectories of various moving objects in the object space.
In this study, we use two PTZ cameras to track the moving objects and apply the photogrammetric technique to position the moving objects in the object space. The proposed method involves two main stages: in the first stage, a camera calibration procedure is employed to construct the collinearity relationship between the image plane and the object space, and then the transformation between the image and objects spaces can be performed; in the second stage, the moving objects on each PTZ camera are detected using the progressive background subtraction method, and the epipolar geometry and space intersection technique are used to calculate the object coordinate of the moving objects. After all, 3-D trajectories of the moving objects can be constructed using the temporal relationship in the image sequence.
The proposed system is tested on various scenarios in an open area. The experimental results reveal that the both of the proposed system and method can successfully track the moving objects and present the 3-D trajectories of the moving objects in the object space.

並列關鍵字

PTZ Camera ； 3-D trajectory ； tracking ； positioning ； moving object ； multiple cameras

參考文獻

王俊明、陳世旺，2002。「漸進背景影像的建構」，師大學報：數理與科技類，第四十七卷，第二期，43-54。

劉虹妤、曾義星、王聖鐸，2001。「數位相機檢定之自動化」，航測及遙測學刊，第六卷，第三期，23-35。

張智安，2002。「EROS A 衛星影像幾何改正之研究」，碩士論文，國立中央大學土木工程研究所。

Brown, D.C., 1971. Calibration of Close Range Cameras, Photogrammetric Engineering, Vol. 37, 855-866.

Khan, S. and M. Shah, 2003. Consistent Labeling of Tracked Objects in Multiple Cameras with Overlapping Fields of View, IEEE Trans. Pattern Analysis and Machine Intelligence, Vol. 25, No. 10, 1355-1360.