::: 前往中央內容區塊
:::

歷史雜誌

說明說明說明說明說明說明說明說明說明說明說明說明

|影像式先進駕駛輔助系統之雨刷雜訊濾除

作者 林哲聰陳隆泰

刊登日期:

摘要:隨著車輛不斷普及,交通事故所造成的死亡人數始終居高不下,先進駕駛輔助系統(Advanced Driver Assistance Systems, ADAS),如車道偏離警示系統(Lane Departure Warning System, LDWS)或前車防碰撞警示系統(Forward Collision Warning System, FCWS)的出現使得行車安全性得以提昇。為達到道路與環境偵測的目的,這類系統應用不同技術類型的感測系統;其中,影像式感測器由於成本相對較低,因此也廣為車廠選用。然而由於影像式感測器普遍裝在車內,因而在下雨時無可避免地會受到來回擺動雨刷的干擾,辨識標的之外觀因此週期性地受到遮蔽,導致辨識率降低與系統可靠度降低。針對此一問題,本文提出一結合影像分割(Image Segmentation)與影像修補(Image In-painting)的雨刷雜訊濾除演算法,其可有效濾除影像中之雨刷物件,實驗結果證實,FCWS使用去除雨刷雜訊後之影像,其前車辨識率確實提昇。

Abstract: Due to rapid increase in the number of vehicles on roads, each year traffic injuries take an enormous toll on individuals and communities as well as on national economies. Recently, ADAS (Advanced Driver Assistance Systems), such as LDWS (Lane Departure Warning Systems) and FCWS (Forward Collision Warning Systems), have been applied widely to help improve driving safety. There are several types of sensors that can be used for road and environment sensing. Among them, camera, relatively cheap, is also used for ADAS by many car makers. However, since the camera is usually installed inside the cabin, it is inevitable that rain wipers would partially and periodically block obstacles on the road that are to be detected in the images being perceived. Thus, the performance of such systems is degraded due to such noise. This paper proposes a wiper elimination algorithm which consists of image segmentation and image in-painting techniques. The experimental results showed that the wiper interference in image could be effectively removed and vehicle detection rate was significantly improved.

關鍵詞:雨刷雜訊濾除、惡劣天候、先進駕駛輔助系統

Keywords:Wiper noise elimination, Adverse weather, Advanced Driver Assistance Systems

前言
據美國高速公路交通安全局(NHTSA, National Highway Traffic Safety Administration)統計,百分之十七的死亡事故,百分之二十二的傷害事故以及百分之二十五的僅有財損(Property Damage Only, PDO) 事故發生在惡劣天候下[1]。近年來,汽車業的一門顯學即是導入各式各樣的先進駕駛輔助系統(Advanced Driver Assistance Systems, ADAS),如車道偏離警示系統(Lane Departure Warning System, LDWS)或前車防碰撞警示系統(Forward Collision Warning System, FCWS)。在良好的天候下,這類系統通常可正常的運作,使得駕駛安全性提昇。然而,在下雨的情形下,這些系統的辨識率即會因能見度降低或障礙物外觀受到遮蔽因而降低。因此,發展抗惡劣天候之影像雜訊濾除演算法是必須的。然而,過往許多泛用型演算法並不適用於不良天候下行車過程之影像,原因是天候對行車影像所造成的雜訊並不是一般訊號傳輸過程或是感測器本身所產生之隨機雜訊。許多研究曾試著提出各類下雨時之影像雜訊濾除演算法,如參考文獻[2-5]提出雨紋濾除演算法,然而,這類演算法只能在靜態背景下運作,因此並不適用於行車環境,最近,相關研究已經進步到可以濾除單張影像中之雨紋[6],這意味著即便是動態背景亦能適用。然而,由於車規攝影機相較於商規,其影像解析度較低,因此,雨紋對影像畫質的影響較為有限。相較之下,雨刷這種往覆遮蔽部份影像像素的雜訊,反而最為顯著。到目前為止,並沒有任何研究分析雨刷對辨識率的影響並提出相關的解決方案。

在行車過程中,一旦雨刷啟動後,前方路面上的車輛外觀即會週期性地部份或完全被遮蔽,如果我們可以分割出雨刷物件並找出相鄰影像中未受雨刷影響的對應像素來修補雨刷物件所佔據之像素,雨刷雜訊即可被濾除。本文所提出的雨刷雜訊濾除演算法由影像分割(Image Segmentation)與影像修補(Image In-painting)所構成。典型受雨刷干擾的影像如圖1(a),本文之目標即是將雨刷雜訊濾除,使其如圖1(b)所示地完全消失,各種障礙物辨識系統基於這樣的影像,都可提昇其辨識率。

更完整的內容歡迎訂購 2014年04月號 (單篇費用:參考材化所定價)

1篇150

NT$150
訂閱送出

3篇450元

NT$450
訂閱送出

10篇1200元

NT$1,200
訂閱送出
主推方案

無限下載/年 5000元

NT$5,000
訂閱送出