視覺化嵌入式交通監控系統-以雪山隧道CCTV影像為例

Abstract

本研究目的是利用影像處理技術來實現交通監控系統。系統由四大單元所構合而成，分別是「自動背景取得與更新」、「自動車道線辨識」、「車輛偵測及追蹤」、以及「事件偵測」。透過自動背景取得與更新單元，系統可以在車流影像中分析出背景並在長時運轉下更新背景，特別是當隧道內出現瞬間燈光切換時，該單元會透過光線校正技術快速地調整背景色彩；車道線辨識單元不論在實線車道線或虛線車道線，皆可以正確地辨識出車道線位置；利用車輛偵測及追蹤單元進行交通影像分析，不僅可以獲得佔有率與交通流量等資訊外，還可以應用於事件偵測單元以偵測出異常停止車輛、違規變換車道、車輛壅塞與煙霧等事件的發生；系統同時記錄事件發生的完整影像，即包含事件發生前之片段；透過事件管理介面的操作，監控人員更可以快速地進行事件的排除。 本系統在高速公路局坪林行控中心內針對雪山隧道既設CCTV 影像進行一個月的長時測試，系統的平均偵測率可達到95%以上。此外，系統不僅可以在個人電腦中執行，自動車流偵測模組更實現於小體積(長8cm 寬6.5cm 高１cm)的嵌入式平台中，未來透過此平台，可輕易地與各地交通控制中心的CCTV 監視系統進行內嵌與系統整合。若與已建置的感應線圈式偵測器相互比較，本系統在功能與建置成本上，皆能取得優勢。

The goal of this study is using image processing technologies to implement the traffic monitoring system. The system is composed of four units which are background image extraction and update unit (BEUU), vehicle detection and tracking unit (VDTU), lane mark recognition unit (LMRU) and incident detection unit (IDU). The system gets an accurate background image and updates it by BEUU despite of vehicle passing through and weather changing. By the way, for the problem of suddenly illumination change, the luminance modification method is proposed to adapt the change; LMRU recognizes the position of lane mark regardless of solid line or dash line; The information provided by VDTU contain not only the traffic parameters but also the trajectories of the vehicles. IDU will reference the given information to detect incidents such as stopped vehicles, illegal lane-changed vehicles, degree of traffic congestion and smoke. Besides, the incident recording procedure is applied to store the full clips of incidents, especially that the period before incident happening is included. With the help of designed software interface, the operating staff will handle the traffic condition much efficiently. The designed system has been one-month verified in Hsueh-Shan Tunnel, and the averaged flow detection ratio is above 95%. Furthermore, the addressed methods are implemented in the embedded platform which owns the advantages of a small volume (8cm x6.5cm x1cm) and a reasonable price. The embedded platform will be easily integrated to the constructed CCTV systems. As a result, the system is much attractive under the consideration of functionality and cost.