本文針對無人駕駛飛機(unmanned aerial vehicle; UAV)提出一影像式之即時興趣區域追蹤技術，結合特徵點估測與追蹤、單應性轉換矩陣估測、卡曼濾波器追蹤及粗差點去除技術，可得到穩定且平滑的興趣區域追蹤。在實驗方面，我們設計了相對應的雙攝影機機台作為模擬情境，也蒐集了實際的無人駕駛飛機空拍影片作測試之用。實驗結果顯示了本方法之準確率、穩定性與運算效率均明顯優於現有其他技術包含mean shift與RANSAC，兼顧了各方面的考量，因此可望廣泛應用到各種無人駕駛飛機監控情境，包括國土安全及救災防護之用。

In this article, we propose an aerial surveillance system utilizing a static camera and a PTZ camera. The static camera captures a wide angle view of the scene, while the PTZ camera is capable of zooming in to a region of interest (ROI) selected by the user and capturing a detailed image. The system consists of a novel tracking algorithm named Simultaneous Estimation, Filtering and Rejection (SEFAR) and a dual-camera apparatus accommodating real-time requirements. The SEFAR algorithm performs feature point detection, homography estimation and outlier rejection in a one-pass process, avoiding time consuming iterations. The multi-camera apparatus uses high-resolution cameras of different zoom factors with a rotatable mirror for faster PTZ movements. Experimental results show robust and efficient tracking performance of SEFAR, demonstrating its possible use in real aerial surveillance systems in the future.