本論文設計並實現了一套結合廣角與長焦鏡頭、基於深度學習之多攝影機協同監控系統，目的在於同時滿足大範圍偵測與遠距目標細節識別之需求，該系統旨在監視軍事場景中的敵對無人機。 由於在天空飛行的無人機距離遠，在畫面中為一個模糊的移動黑點，偵測模型無法輕易判斷，因此本研究使用一支廣角攝影機以及多支長焦攝影機協同工作。廣角鏡頭能夠即時偵測畫面中的無人機位置，並判斷無人機是否進關注區域，若無人機進入關注區域，則啟動該區域對應的長焦鏡頭，以擷取更高解析度的影像，而其餘閒置之長焦鏡頭則不啟動以降低系統運算負擔。在本系統中透過訓練好的YOLOv2無人機檢測器辨識目標位置，將長焦無人機部分裁切、放大疊回廣角影像當中，以達到「廣域監控+局部細節放大」之功能。 本系統具有結構簡潔、運算效率高、易於擴展的特點，適用於安防監控、場域偵測、智慧交通等需求遠距目標細節觀測之場景，為多攝影機協同應用與深度學習監控技術之整合提供新穎且高效的解決方案。 ;This study designs and implements a multi-camera cooperative surveillance system that integrates wide-angle and telephoto lenses, based on deep learning, to achieve both wide-area detection and detailed identification of long-distance targets. The system is intended for monitoring hostile drones in military environments.. Due to the long distance of flying drones, they often appear as blurry moving black spots in images, making them difficult for detection models to accurately identify. To address this, the proposed system employs a wide-angle camera in coordination with multiple telephoto cameras. The wide-angle camera performs real-time monitoring to locate potential drone positions and determine whether they enter predefined areas of interest (AOIs). Once a drone enters an AOI, the corresponding telephoto camera is activated to capture high-resolution imagery, while other telephoto cameras remain idle to reduce computational load. A trained YOLOv2 drone detector is used to identify targets, and high-resolution images captured by telephoto lenses are cropped, enlarged, and overlaid onto the wide-angle view, achieving the goal of "wide-area monitoring with local detail enhancement." The system features a simple architecture, high processing efficiency, and scalability. It is well-suited for applications requiring long-distance object recognition, such as security surveillance, field monitoring, and intelligent transportation. This work offers a novel and efficient solution for integrating multi-camera systems with deep learning-based monitoring.
