利用多支攝影機連接監視區域以進行車輛偵測之研究

Abstract

在本論文中，我們使用單支監視攝影機擷取道路影像，針對有興趣之區域（ROI）利用路面色度模型(chromaticity color model)分離出影像中路面與非路面之圖素，然後針對非路面圖素之集聚區域來進行車輛偵測之判別，同時計算出道路上之總車輛數，以提供有效的車輛密度、車行流量之交通參數作為交通控制管理之用。但由於單支監視攝影機所監視之區域範圍相當有限，對於某些較長遠距離範圍之重要交通路段就只能偵測其部分之車流情況，並未能獲得完整路段之車流資訊。因此，我們發展一套使用多支監視攝影機擷取道路影像，根據平坦、上坡、下坡三種不同的道路路面幾何狀況及攝影機之姿態，計算出單支監視攝影機所監視範圍之最短距離來連接相鄰兩支監視攝影機所監控之區域，以擴展攝影機所監控之區域範圍，提供更完整之車流資訊並且有效地控管整體路段之交通狀況。

In this thesis, the road images are captured by a single surveillance camera, and the chromaticity color model of road is used to filter each pixel within region of interest (ROI) as a road or non-road. Then non-road pixels will be grouped into regions and be identified as vehicles or not in the image. The total number of vehicles on the real road plane can be obtained to offer the vehicle density and flow rate for traffic control and management. The inspected range is often limited by using a single surveillance camera. Hence the proposed approach can extend inspected range of a single surveillance camera wider with multiple surveillance cameras. The rules of extend inspected range can be achieved by calculating the minimum distance of field of camera’s view according to cameras pose and the road geometry of flat road plane, upward road plane and downward road plane. The proposed approach is proven successful to combine two contiguous inspected regions, and can provide more completely traffic flow information for traffic control and management.