視覺式倒車導引與障礙物偵測系統;Vision-based backward parking guidance and obstacle detection system

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Computer Science and Information Engineering > Electronic Thesis & Dissertation > Item 987654321/79583

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/79583

Title:	視覺式倒車導引與障礙物偵測系統;Vision-based backward parking guidance and obstacle detection system
Authors:	林昱岐;Lin, Yu-Chi
Contributors:	資訊工程學系
Keywords:	倒車導引;障礙物偵測;移動估計;parking guiding;obstacle detection;motion estimation
Date:	2019-01-31
Issue Date:	2019-04-02 15:04:02 (UTC+8)
Publisher:	國立中央大學
Abstract:	近年來，車輛俯瞰監視(top-view around monitor)系統已經成為一個標準配備的駕駛輔助工具，藉此減少因盲點造成的碰撞危險。許多此類系統提供車輛周圍的環場影像，適用於擁擠空間的行駛與倒車停車。美國國家公路交通安全管理局已經發佈了規定，2018年5月後所有低於一萬英磅的新車輛都必須搭載車後影像設備。車後相機已經成為了未來的標準備配。在本論文中，我們提出了一個倒車導引與障礙物偵測系統，只需要一部車後廣角相機無須搭載其他感應器，便可在車後影像，產生俯瞰影像及倒車導引線，並標記障礙物以提醒駕駛人。在倒車導引的部分，我們先偵測影像中的特徵點，透過絕對誤差和數值（SAD）來匹配前後時刻影像中的特徵點。每對成功匹配的特徵點都可得出對應的運動向量，再使用最小平方誤差估計法估計車輛的運動參數。基於阿克曼轉向幾何模型，車輛的運動參數可估計出車輛的運動軌跡，並根據車輛運動軌跡繪製停車導引線。在障礙物偵測部分，透過已經得到的連續影像及估算出的車體自我運動向量，系統便可以比對偵測到特徵點與地面的運動是否相似，藉此初步篩選出不是躺平在地面的物件為候選障礙物，再透過表面法線估計評估各物件與地面法線的夾角，來確認物件為高於地面的立體物或是平躺於地面不會造成威脅的標線或圖案。以上方法在大多數情況都可以有效偵測障礙物，但偶而會受到相機震動或影像雜訊干擾。倒車導引和障礙物偵測是一個連續的動作，若不良影像只是少數幾張，則可以利用機率來排除。當系統判斷該點可能為障礙物便將該點與鄰近點增加計數，將計數累積成熱度圖後，機率高的部分視為障礙物，減少因為輸入瑕疵造成的誤判。最後根據駕駛者的喜好可於俯瞰影像或是原始影像中標記偵測到的障礙物，以提醒駕駛者潛在的危險。 ;In recent years, top-view monitoring systems are becoming a practical driving aid that help reducing collision hazards by eliminating blind spots. The U.S. Department of Transportation’s National Highway Traffic Safety Administration (NHTSA) issued a rule requiring rear visibility technology in all new vehicles under 10,000 pounds by May 2018. Many of such systems provide short range views surrounding the vehicle, limiting its application to parking and reversing. In this paper, we propose a practical system for creating top-view image of the vehicle with the parking guidance line, and highlighting obstacles only relies on a wide-angle camera to capture images for analysis without sensors. In the proposed parking guidance system, feature points on two consecutive images are extracted to match each other. First, the feature-point pairs are further pruned by Sum of Absolute Differences (SAD)。The remained pairs are then used to estimate vehicle motion parameters by a least-square error metrics, where an isometric transformation model based on the Ackermann steering geometry is proposed to describe the vehicle motion. At last, the vehicle trajectory is estimated based on the vehicle motion parameters and the parking guidance lines are drawn according to the vehicle trajectory. In the proposed obstacle detection system, by estimating the ego-motion of the vehicle using the input image sequence of the cameras, the system is able to detect objects in the images by finding movements of features that do not correspond to ground motion relative to vehicle motion. Then confirm it by the surface normal estimation for the angle of object and ground. Parking guidance and obstacle detection are a continuous action, excluding error by probabilities. Increase the count in heat map when the object is detected, the object with high counts is marked as an obstacles, which are highlighted in the multi-view imagery to warn the driver of potential hazards.
Appears in Collections:	[Graduate Institute of Computer Science and Information Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	166	View/Open

社群 sharing

Loading...