3D 環境中具有時域涵蓋能力之循環式攝影機佈建及排程設計

Abstract

隨著安全議題逐漸升溫，環境監控日漸受到重視，無論在居家社區、公司校園及工廠環境等，都需要一套良好的監控系統來維護人員生命及財產的安全。然而，傳統的攝影機監控系統主要採用固定式的攝影機，當環境中存在多個目標物需要被監控時，如：出入口、開放性走道區域或私人空間等，若要達成全面性的監控，則需在每個地點獨立佈建固定式攝影機，如此將大幅提高佈建成本，同時產生大量不必要的影像資源浪費。因此，利用具有水平/垂直旋轉及變焦能力之攝影機(Pan-Tilt-Zoom, PTZ)，是未來監控系統的重要趨勢之一。然而，如何發揮PTZ攝影機之間的協同合作，同時確保監控目標物之完整的監控品質條件(如：可視角度、影像解析度及時域特性)，再透過妥善的排程規畫設計，以有效減少攝影機的佈建成本，達成具有循環監控能力的智慧型監控系統，將是一項新的挑戰。在本篇論文中，我們針對以上目標，提出了一套高效且低複雜度的PTZ攝影機佈建及循環排程演算法，其共分為三個階段。在第一階段中，我們根據攝影機覆蓋物體的能力，以決定可能佈建攝影機之位置。第二階段中，我們考量監控目標之特殊需求條件，找出佈建位置的視野範圍與監控目標的配對集合，以決定合適的攝影機擺放角度。最後，在第三階段中，我們使用最佳剩餘時間策略來進行攝影機排程，並考量監控時效需求及涵蓋物體數量，以最佳化排程結果，同時確保監控品質。根據模擬結果，其驗證了我們的方法的確擁有近似最佳解的監控效益，且運算時間只需要最佳解的百分之一。除此之外，我們還進一步的實作此循環監控之雛型系統，並實地佈建於校內系館實驗室及走道空間，以監控實驗室環境及系館出入空間，除了驗證此系統之有效性外，更提供校園師生之安全及保障。

Environmental surveillance has many applications in households, offices, schools, and factories. Static cameras are commonly used in traditional surveillance systems. However, to monitor multiple targets, the number of a static cameras needed also increases proportionally. This causes high deployment cost and network resources. Therefore, using pan-tilt-zoom (PTZ) cameras, which can rotate horizontally and vertically and change focal lengths, is desired. In this work, we consider (1) how to deploy cooperative PTZ cameras for ensuring monitoring quality requirements, such as view-angle, image resolution, and temporal coverage, and (2) how to schedule these cameras to meet these quality requirements. We propose low-complexity deployment and cyclic scheduling scheme to solve these problems. This scheme consists of three phases. The first phase chooses a position that may cover the largest number of targets. The second phase finds the maximum coverage FoV set and the relationship sets between FoVs and covered targets considering the monitoring requirements. Based on the FoV set, it determines the corresponding camera parameters at the chosen position. The third phase places a PTZ camera at the position and schedules its duty cycle using a Minimal Remain-time Cost First strategy, which considers the surveillance requirements constraints. These phases then repeated iteratively. Simulation results show that our approach is close to the optimum solution and the computation time is less than the optimum 100 times. Furthermore, we have implemented a prototype of this cyclic surveillance system. By real deployment, we show that the system has the ability to provide excellent surveillance.