車輛影片中車道線之偵測及追蹤

Abstract

近年來為了減少交通事故而發展的車輛輔助安全駕駛議題越來越受重視。其中，車道線偵測在車輛輔助安全系統中是一項必需的元件。在這篇論文中，我們提出一個利用time-slice images來進行偵測及追蹤車道線的系統。另一方面，由於虛線較為稀疏，不容易在畫面上偵測到，我們提出梯度值調整方法來提高虛線在畫面上的辨識度。每一張影像經過前置處理、邊緣偵測、峰點找尋及連結找出候選的車道線，接下來利用time-slice images進行車道線驗證，找出真正的車道線位置，最後在time-slice images上預測這些車道線在新畫面的可能位置，以進行進一步的追蹤。由於我們只針對影像中某幾行進行處理，且我們利用追蹤的方法提供了連續的車道線偵測，這會減少一張影像所需要的處理時間。 實驗中採用行車紀錄器所拍下的影片當作測試資料，對於三種情況下做車道線偵測：一是當從直線開到曲線時，二是當進行車道切換時，三是開在直線上時。我們實作相關研究之車道線偵測演算法並對所執行出來的結果及產生的問題進行討論。實驗結果顯示，我們提出的方法確實能協助某些情況下車道線的偵測。

In recent years, in order to reduce traffic accidents, developing Driver Assistance Systems for safety has attracted much attention. Lane line detection is an essential component of Driver Assistance System. In this thesis, we propose a lane line detection and tracking system utilizing the time slice images. On the other hand, due to the discontinuousness, dashed lane lines are difficult to detect in a single image. We propose the gradient value adjustment method to enhance the recognition of the dashed lane lines. For each frame, we find the candidate lane lines in the image through pre-processing, edge detection, and peak finding and connecting. Then we detect the true lane line positions by the time slice images in lane line verification. Once the lane line positions are located, we predict their new positions by the time slice images and track them in the new frame. Since we only process several rows of an image and we provide the continuous detection of the lane lines by tracking, the processing time of an image is reduced. The experiments use the video sequences of real road scenes containing three conditions: (1) the intermediate case of driving from the straight lane lines to curve lane lines, (2) the lane changing case, and (3) the straight lane lines case. We implement other lane line detection algorithms on the above cases, and then analyze the experimental results and discuss the problems arising in the experiment. However, the experimental results show that our proposed methods can improve the lane line detection in some cases.