即時交通影像串流(如:路側攝影、空拍攝影)已成為現今交通監控的主要方法之一,但如何有彈性的整合廣域的監控系統並將影像資料數據化呈現以有效管理,則是一個待解決的問題。在本研究中,我們在稱為SAGE2的開源群組視訊共享平台上建置一套智慧交通監測系統,SAGE2的電視牆提供多視窗資訊呈現的能力,電子地圖做為串流影像來源的管理介面,以空拍機交通影像為資料來源,利用深度學習技術將車流影像換轉為微觀車流數據,透過DL (Deep Learning) as a Service技術提供可彈性調用的深度學習服務以有效整合利用GPU叢集。這個雛型系統目前是以空拍影像為設想,未來可進一步整合已廣泛布建的路側攝影,以提高影像來源的多樣性。
Nowadays real time traffic video streaming (like roadside surveillance and aerial video) has been widely used in traffic monitoring, however, how to flexibly apply a widely distributed monitoring system and video data for effective management is a problem to be solved. In this work, we developed an intelligent traffic monitoring system on an open source group video sharing platform called SAGE2. Based on the integrated big screen TV wall of SAGE2, we designed a map-based aerial traffic video streaming management interface. Combined with Deep Learning as a Service (DLaaS) GPU cluster, our system can provide flexible and intensive deep learning services. It uses deep learning technology to capture microscopic traffic flow data from traffic flow video and use visualized charts to provide to effectively support decisions making. This prototype system is currently based on aerial video. In the future, it can further integrate the wide-ranging roadside surveillance to improve the coverage of monitoring.