標題: 以天花板上多環場攝影機輔助自動車作室內安全監控之研究
A Study on Indoor Security Surveillance by Vision-based Autonomous Vehicle With Omni-cameras on House Ceiling
作者: 王建元
Wang, Jian-Yuan
蔡文祥
Tsai, Wen-Hsiang
多媒體工程研究所
關鍵字: 視覺自動車;安全巡邏;車輛定位;上視攝影機;交棒;vision-based autonomous vehicle;security patrolling;vehicle location;top-view cameras;handoff
公開日期: 2008
摘要: 本論文提出了一個以天花板上多環場攝影機輔助自動車做室內安全監控之方法。我們使用一部以無線網路傳遞控制訊號之自動車,並在其上裝置一部攝影機,用以監視室內環境及拍攝入侵者影像。另外使用兩部裝置在天花板上的魚眼攝影機協助此自動車做導航。在此研究中,我們提出了一種即時對環境中空地和障礙物做定位的技術,並以計算出的位置建立完整的環境地圖及規劃自動車的巡邏路線,使自動車可以在複雜的環境中導航並避開障礙物和牆壁。此外,我們亦提出了一種能夠適應高度變化的空間對映法,利用推導出的公式計算設置在不同高度的魚眼攝影機的對映表,以此對映表配合內插法,為環境中的物體進行定位。因為自動車行走時會產生機械誤差,我們也提出了四個策略,以修正自動車的位置及方向。此外,我們亦提出了一種追蹤入侵者的技術,使用追踨視窗準確地預測及計算人物在扭曲影像中的位置,並在追踨過程中同時記錄人物的特徵。為了擴大可監視的範圍,我們使用了多台裝置於天花板上的攝影機,為此我們也提出了一種在多台攝影機下“交棒” (handoff)的技術,使自動車或入侵者從一台攝影機的視野範圍移動到另外一台時,能夠不間斷的被追蹤。實驗結果證明我們所提出的方法是可行而且有效的。
Vision-based methods for security surveillance using an autonomous vehicle with fixed fish-eye cameras on ceilings in an indoor environment are proposed. An autonomous vehicle controllable by wireless communication and equipped with a camera is used as a test bed and navigates in a room space under the surveillance of multiple fisheye cameras affixed on the ceiling. To learn the information of the unknown room environment in advance, a method is proposed for locating the ground regions, identifying the positions of obstacles, and planning the patrolling paths. The data obtained by the method enable the vehicle to navigate in the complicated room space without collisions with obstacles and walls. Also, a height-adaptive space mapping method is proposed, in which the coordinates of corresponding points in 2-D images and 3-D global spaces are computed by interpolation to form a space mapping table for object localization. Appropriate equations are derived to adapt the table to fish-eye cameras affixed to different ceiling heights. Because the vehicle suffers from mechanic errors, a vehicle location and direction correction method is proposed for correcting the errors according to four strategies. Furthermore, a method for detecting and tracking intruding people is proposed. The approximate position of the person can be predicted first, and the exact position is then calculated via a tracking window in images. Some useful features of the intruding person are computed for person identification. To enlarge the area under surveillance using multiple cameras, the camera handoff problem is also solved by using information of the overlapping regions of the cameras’ fields of view. Experiments for measuring the precisions of the proposed methods and tracking intruding persons were conducted with good results proving the feasibility of the proposed methods.
URI: http://140.113.39.130/cdrfb3/record/nctu/#GT079657536
http://hdl.handle.net/11536/43542
Appears in Collections:Thesis


Files in This Item:

  1. 753601.pdf