以霍夫轉換為基礎之智慧型快速車道線偵測

Abstract

車道線偵測在自動化駕駛佔有關鍵性的角色。在車道線偵測系統中， 為了減少非車道線物件的干擾，通常需要耗費非常大的計算量。此外，傳 統上利用霍夫轉換來偵測車道線，亦因繁重的計算量而降低其實用性。 本論文針對上述的問題提出一個新的解決方案。首先，將偵測範圍縮 小在靠近車輛的區域，以減少非車道線的物件干擾；再者，將車道線以直 線近似；接著透過訂定適當初始條件，以及利用最小平方誤差法，來得到 道路線的斜率；最後，利用霍夫轉換搭配直線方程式，來獲得車道線的位 置。 經由真實道路行駛所錄製的影片驗證，在特定條件如不同天候及震動 幅度較大的狀況下，可穩定且正確的偵測出主車道線的位置。此外在執行 速度方面，每張 640 × 480 的畫面平均只需 17 ms 即可算出車道線位置。

Lane detection is crucial for autonomous driving. In lane detection system, to reduce interferences from non-lane marking objects, it costs a great amount of computations. Moreover, the heavy computations would cause less practicability when applying commonly used Hough transform in lane detection. A new solution of above problems is proposed in this thesis. First, the detection region is narrowed down to regions close to vehicles. This method can reduce the interferences of non-lane marking objects. Second, lane markings are approximated to straight lines for computing simplicity. Third, the slope of lane line is acquired by giving proper initial conditions and computing with the least squared error method. Finally, by applying the above information into Hough transform and linear equations, the position of lane line is located. Examined through actual on-road video, even under specific conditions such as different weathers and greater bumping, the proposed method can steadily and correctly detect the lane lines. Moreover, in executing efficiency, only 17 ms is needed to calculate out the line position in a 640 × 480 frame.