遍佈式路面異常狀況監控與旋轉角度調整

Abstract

近年來，由於汽車的數量大增，使得行車安全議題越來越受到重視。在用路安全方面，路面異常的議題在近幾年備受關注，路面品質不良不僅會造成車輛的損壞，甚至會造成交通事故的產生。道路品質的維護一直以來都是一個難解的問題需要耗費大量的人力及時間，主要是因為各種大大小小的道路涵蓋面積龐大、範圍廣，以往週期性的人工檢測方式往往無法即時地監控路面異常的狀況，因而造成意外發生。行車紀錄器在近年越來越普及化，驅使了許多學者投入相關的研究，行車紀錄器因其遍佈性及移動性，非常適合用來監控道路表面的狀況。 本論文提出了遍佈式路面異常狀況監控與旋轉角度調正，使用了遍佈的行車紀錄器當作感測器，用以監控路面異常的狀況。一開始，我們分析行車紀錄器的影像畫面藉此偵測車輛是否行經路面異常的區域，若是有，則會將路面異常狀況與GPS位置回報至中央路況監控伺服器進行後續的分析。因此，往後的駕駛人可以依照這些訊息選擇路面狀況較佳的路段行駛。除此之外，這些訊息也能提供給相關的單位做路面維護的參考。 另外，本論文也將此偵測方法套用到機車上，裝置在機車上的攝影機所錄到的影像與汽車行車紀錄器最大的不同在於多了旋轉角度的變化，本論文估算出機車的傾斜角度後將影像調正，再進行後續的路面異常偵測。

Nowadays, advanced traffic safety is an emerging issue appealed from the rapidly explosive population of vehicles. Road anomalies not only damage vehicle but may also cause accident threatening the drivers, especially under poor visibility condition. However, maintaining the quality of road surface is a time-consuming task that has been a big challenge. Recently, the widespread popularity of driving video recorders (DVRs) makes it possible to be a distributed sensor to detect event along the road. In this thesis we propose pervasive vision-based road anomaly monitoring with roll angle compensation, we employ the widespread DVRs equipped on vehicle as distributed sensor to detect road anomalies with high mobility. First, vehicle vibrations are detected to infer candidate anomalies. Then the candidates will be validated in a monitoring server and stored in a database. For a video recorded by a camera equipped on a scooter, the roll angle will be compensated in a pre-processing step. With the locations of anomalies, other drivers can receive warnings when approaching detected anomaly regions and maintenance community can repair the road surface. Experiments are conducted on a test data set collected by front-mounted DVRs and HTC RE Camera, and the experimental results verify that the proposed system can detect road anomalies efficiently and effectively, showing its feasibility in real world.