本研究設計並開發了一套整合廣角鏡頭與長焦鏡頭的多鏡頭協同監控系統，並導入深度學習技術，旨在同時兼顧大範圍場域偵測與遠距離目標的細節識別。本系統特別針對軍事防禦場景中的敵對無人機監控進行優化。 鑑於空中無人機因距離遙遠，在影像中常呈現為模糊且難以辨識的移動微小暗點，傳統偵測模型不易精準判斷。對此，本系統採用「單支廣角鏡頭」搭配「二十支長焦鏡頭」的架構：廣角鏡頭負責即時捕捉全景，全時監控無人機方位並判定其是否進入預設之關注區域；一旦目標侵入特定區域，系統將精準觸發對應的長焦鏡頭進行局部高解析度影像擷取，其餘未受影響之長焦鏡頭則保持休眠，以大幅降低整體系統的運算負擔。在核心演算法方面，本系統利用訓練完成的 YOLOv8物件偵測模型精確定位目標，並將長焦鏡頭所拍攝之無人機特寫進行裁切與放大，隨後即時貼回廣角全景影像中，完美實現「廣域監控與局部細節兼得」之雙重效益。 本系統具備架構簡潔、運算效率卓越及高擴展性等優勢，除軍事國防外，亦可延伸應用於安全防禦監控、特定場域偵測及智慧交通等需遠距精細觀測之領域，為多監控鏡頭協同作業與深度學習技術的整合應用，提供了一套創新且高效的解決方案。;This study designs and develops a multi-camera collaborative surveillance system that integrates wide-angle and telephoto lenses， enhanced by deep learning technology， to achieve both wide-area detection and high-resolution detail identification of long-range targets. The system is specifically optimized for countering hostile drone threats in military defense scenarios. Given that airborne drones at long distances often appear as blurry， faint， and minuscule moving dots in images， conventional detection models frequently struggle with precise identification. To address this challenge， the proposed system adopts a hardware architecture comprising a single wide-angle camera paired with twenty telephoto cameras. The wide-angle camera is responsible for real-time panoramic tracking， continuously monitoring the drone′s bearing and determining whether it has entered a predefined area of interest. Once a target breaches the specified zone， the system selectively triggers the corresponding telephoto camera to execute a high-resolution localized capture， while the remaining inactive telephoto lenses remain in sleep mode， thereby drastically reducing the overall computational overhead. In terms of the core algorithm， the system utilizes a trained YOLOv8 object detection model to precisely localize targets. The close-up drone captures obtained by the telephoto lenses are cropped， enlarged， and seamlessly stitched back onto the wide-angle panoramic frame in real time， successfully achieving the dual benefits of expansive surveillance coverage and meticulous localized detail. The developed system boasts a streamlined architecture， exceptional computational efficiency， and high scalability. Beyond military and defense applications， it can be readily extended to security surveillance， restricted airspace monitoring， and intelligent transportation systems that demand precise long-range observation. This work offers an innovative and highly efficient solution for the integration of multi-camera collaborative operations and deep learning technologies.
