影像特徵匹配用於室內定位;Image Feature Point Matching for Indoor Positioning

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Computer Science and Information Engineering > Electronic Thesis & Dissertation > Item 987654321/73011

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/73011

Title:	影像特徵匹配用於室內定位;Image Feature Point Matching for Indoor Positioning
Authors:	李祁晉;LEE, CHI-JIN
Contributors:	資訊工程學系
Keywords:	SIFT;影像配對;K-D樹;FLANN;SIFT;image matching;K-D tree;FLANN
Date:	2017-01-23
Issue Date:	2017-05-05 17:38:29 (UTC+8)
Publisher:	國立中央大學
Abstract:	隨著科技的日新月異與智慧型手機的普及，人們對於身心障礙人士的生活便利性考量，增加許多輔助設備，方便攜帶的設備例如智慧型手機或平板電腦是每人生活必備，但應用程式較少對視障人士或是老年人的方便性做考量。本系統應用在室內環境得知使用者當下位置並規劃路線導航，透過影像特徵匹配技術去確認出正確位置與方向，目的是幫助視障人士或是老年人提供協助，導航到目的地，另外也可以使用在機器人的定位方式，以影像的方式去做位置與方向的定位。本系統提出使用智慧型手機鏡頭擷取影像並從資料庫中去比對出影像最為相似的地點，傳回使用者目前的位置，使用者可以簡單地得知位置附近有用的資訊，例如廁所位置或是電梯位置，協助使用者規劃路線。本論文使用Scale-invariant feature transform(SIFT)特徵，藉由影像特徵與預先建立完成的場景影像特徵資料庫來做匹配，使用FLANN(Fast Library for Approximate Nearest Neighbors)建立多棵隨機K-D樹(Randomized K-D tree)，K-D樹可以將SIFT特徵描述子建立索引的方式加快特徵配對的效率。經實驗結果，本論文使用的方法可以達到良好的可靠度。 ;With the development of technology and the popularity of smart phones, people have become more considerate about the convenience in life of the disabilities and thus add lots of assistive equipment. Portable equipment such as smart phones and tablet are daily necessities for most of the people. However, there are few applications that are designed for the visual impairments and the elders. The proposed system can be used indoor to get the location of the user and plan the path. Via matching the image features, it will recognize where the user is. The purpose of this is to help the visual impairments or the elders to navigate to their destinations. In addition, it can also use robot to get the position by matching image features. The proposed system uses images captured from smart phones and are compared with database, then it will return the most similar position back to the user. User then can obtain some useful information of the surroundings such as the location of toilet or elevator, which will help user planning the path to destinations. Scale-invariant feature transform(SIFT) is used in this thesis. Via image features matching with pre-established scenes image features database. FLANN(Fast Library for Approximate Nearest Neighbors) is applied to build randomized k-d trees. The k-d tree can create an index for SIFT`s descriptor, which can speed up feature matching. Experimental results in the proposed method can achieve a good feasibility.
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	380	View/Open

社群 sharing

Loading...