CASIA OpenIR  > 毕业生  > 硕士学位论文
Thesis Advisor高伟
Degree Grantor中国科学院研究生院
Place of Conferral北京
Keyword基于图像的相机定位 运动恢复结构 点云模型
Other Abstract
基于图像的相机定位是三维计算机视觉研究中的一个基本问题,其任务是根据相机拍摄的单幅图像来计算相机在场景中的位姿,包括相机的位置和姿态。本文探索了面向室外场景的移动端位姿估计技术,采用客户端/服务器模式,通过手机端拍摄图像,然后使用BRISK(Binary Robust Invariant Scalable Keypoints)方法,进行二维局部特征提取和描述,最终在客户端进行相机图像的快速位姿估计。主要贡献如下:
第一,本文详细描述并分析了已有的四种运动恢复结构(Structure from Motion, SfM)开源软件包:Bundler、VSFM、COLMAP、openMVG,然后通过实验比较各自的优缺点,最终选出模型重建精度最好的方法进行场景三维重建。实验结果表明,COLMAP具有最好的精度,VSFM具有最快的重建速度。本文在对图像进行位姿定位的过程中,场景三维模型的精度是确保定位结果准确的前提。所以,本文将采用COLMAP方法进行场景三维重建。
第二,本文提出了一种有效获取场景的BRISK点云模型的方法和流程。通过实验发现直接使用BRISK描述子进行特征匹配并三角化,无法得到准确的三维点云模型和相机姿态,因为BRISK的描述性能没有SIFT(Scale-Invariant Feature Transform)好。本文提出两步重建流程来获得场景的BRISK点云模型。第一步使用COLMAP的方法采用SIFT特征提取与描述进行场景的三维点云重建和相机位姿估计,第二步利用已估计的相机位姿,进行BRISK特征提取与引导匹配,然后通过三角化得到BRISK点云模型。实验结果表明,通过两步重建方法可以在内点率相近的情况下,重建获得更多的BRISK三维点。
Image-Based Localization (IBL) is an essential task in the 3D computer vision area. It aims at computing the camera pose, including position and orientation, with respect to the given geometric model of a scene from a single image. This thesis implements an image-based localization application based on the mobile phone for the outdoor scene. The mobile phone and a server work in client/server mode, thus the mobile phone captures a query picture and sends it to the server to compute the pose of the mobile phone. It accelerates extracting and computing local descriptors of the query image by adopting binary local descriptor BRISK (Binary Robust Invariant Scalable Keypoints) and achieves fast camera localization. Out main contributions are:
First, this thesis presents a comprehensive description and study of four different Structure from Motion (SfM) software packages: Bundler, VSFM, COLMAP, and openMVG, and compares theirs merits and demerits through extensive experiments. The purpose is to evaluate the mapping ability of each method and to select the most accurate one to reconstruct BRISK point cloud model (PCM). The experiment results show that COLMAP has the best accurateness while VSFM the fastest speed. Accurateness of the PCM is the precondition for localizing effectively and accurately, so we will adopt  COLMAP in this thesis.
Second, this thesis proposes an effective and efficient pipeline of acquiring the BRISK PCM. It is found that directly matching BRISK descriptors and triangulating pipeline will lead to inaccurate PCMs and camera poses, since the BRISK descriptor is not as robust as the SIFT (Scale-Invariant Feature Transform). This thesis will introduce a two-step reconstruction pipeline to acquire a BRISK PCM. Thus, firstly acquiring a SIFT PCM using COLMAP and secondly matching BRISK descriptors based on the known camera poses, and finally triangulating to acquire a BRISK PCM.
Third, this thesis implements a fast camera localization system based on mobile end. We adopt client/server mode to relieve the limitation of computing resources of the mobile platform. On the server side, establishing the correspondences between local features of a query image and 3D cloud points with related feature descriptors used during the reconstruction through matching algorithm and then robustly estimating
camera pose using Perspective-n-Point (PnP) algorithm. On the mobile side, detecting and computing BRISK feature descriptors for each captured query image, and then sending it to the server for computing camera pose and finally fetching the result.
Document Type学位论文
First Author AffilicationInstitute of Automation, Chinese Academy of Sciences
Recommended Citation
GB/T 7714
晏超超. 面向室外场景的移动端位姿定位技术[D]. 北京. 中国科学院研究生院,2017.
Files in This Item:
File Name/Size DocType Version Access License
面向室外场景的移动端位姿定位技术.pdf(14506KB)学位论文 暂不开放CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[晏超超]'s Articles
Baidu academic
Similar articles in Baidu academic
[晏超超]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[晏超超]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.