基於賈柏濾波器及色彩資訊進行車道線偵測

近年來為了避免因為駕駛者沒有專心或是錯誤的判斷所導致的車禍意外，駕駛輔助系統成為十分熱門的議題。車道線偵測在駕駛輔助系統中佔有關鍵性的角色，能在車道偏離時給予警示或是能夠自動修正有危險的動作。本篇論文提出一個基於紋理的方法，從單眼相機得到的影像中，利用賈伯濾波器來找尋影像中的紋理，並且基於對車道線的三個假設，結合色彩資訊和紋理分布，篩選出可能是車道線的區域，並利用直線模型來表示其對應位置。本方法主要分成三個部分，首先是利用賈柏濾波器找尋紋理，藉由影像中的紋理找到消失點位置，並結合車道線的色彩資訊，去除不相關的地方。再來，利用紋理強度累加的概念，找到具代表性的車道線紋理方向，最後將符合車道線紋理的區域表示出來，並藉由紋理一致性來最佳化代表車道線的直線。本文的車道線偵測分別在高速公路、城市、鄉鎮、校園的不同場景做實驗，在多數情況下車道線可以被準確偵測，但在紅線過暗或是車道線被障礙物遮蔽時，偵測會發生錯誤，基於兩個條件判斷，在高速公路上能達到較理想的96.67% 成功率，其他如城市、鄉鎮、校園則只能分別達到 60%, 80%, 50%的成功率。

關鍵字

賈柏濾波器；紋理偵測；消失點偵測；車道線偵測；單眼相機

並列摘要

In recent years, in order to avoid the accidents which are caused by the lack of recognition and miss judgement by drivers, the driving assistance system become a very popular issue. Lane detection plays an important role in the driving assistance system. The position of lane can provide information for system to give a warning when the vehicle occur departure or automatically correct the dangerous action in the future. This paper presents a texture-based method. Gabor filter is utilized to find the texture orientation on the image from monocular camera. Based on the three assumptions on the lane, combine with color cue and texture orientation distribution to select the possible lane region. At the end, use the line model to represent the lane region. The method is divided into three parts, the first is to find the texture by Gabor filter, and detect the vanishing point with known texture orientation, then remove the unrelated region with the color information of lanes and detected vanishing point. After that, find the representative lane texture orientation with the accumulation of texture intensity. At the end, mark the region in lane texture orientation, and optimize the representative line with orientation consistent number (OCN). In this thesis, the lane can be detected accurately in the difference scenes, including freeway, downtown, country, and campus. In most time, the lane can be detected accurately except the red lane is too dark or the lane is cover by objects. As a result, based on two conditional limitations, the experiment in freeway can reach 96.67% successful ratio, other scene such as downtown, country, campus can reach 60%, 80%, and 50% respectively.

並列關鍵字

Gabor filter ； vanishing point detection ； texture detection ； lane detection ； monocular camera

參考文獻

[1: Borkar et al. 2012]

Amol Borkar, Monson Hayes, Mark T. Smith, “A Novel Lane Detection System With Efficient Ground Truth Generation,” IEEE Transactions on Intelligent Transportation systems, VOL. 13, NO. 1, pp. 365 – 374, Mar. 2012.

[2: Wu et al. 2015]

Joon Woong Lee, “A Machine Vision System for Lane-Departure Detection,”

Computer Vision and Image Understanding, VOL. 86, NO. 1, pp. 52-78, Apr. 2002.

國際替代計量

基於賈柏濾波器及色彩資訊進行車道線偵測

全文下載

主題瀏覽