使用三維全景重建法進行電腦視覺導航

Computer Vision Based Navigation with 3D Scene Reconstruction

李柏儀(Po-Yi Lee)

指導教授 ： 蕭富元

淡江大學/工學院/航空太空工程學系碩士班/碩士(2016年)

https://doi.org/10.6846/TKU.2016.01028

摘要

本論文主要探討利用三維全景建構的方式，來進行無人機的電腦視覺導航。傳統上無人機大多使用GPS 與中途點的方式來進行導航，但GPS 訊號在室內容易因為被建築物擋住而接收不到。所以在本研究中，利用建構無人機四周的三維空間資訊，找出空間中“最深”的點，並假設該點應該是走道或是任何開口，因而可以用來探索整棟建築物裡面的空間。研究時，我們比較不同的三維全景建構演算法，並尋找出空間中最深處的點，不斷透過影像處理的基礎方法提升準度及效率。在三維全景建構模擬與室內通道行走的實驗來展示本研究的可行性。本論文的成果將來可擴充無人機在室內的應用性。

關鍵字

立體視覺 ； 影像處理 ； SSD 演算法 ； 特徵點校正 ； 影像幾何中心校正

並列摘要

This thesis investigates the computer-vision based navigation of an unmanned aerial vehicle (UAV) using 3D scene reconstruction. Conventionally,UAVs are usually navigated with GPS signal and waypoints. This method does not work indoors, since most of GPS signal is usually blocked by buildings. In this research, we intend to navigate the UAV by constructing the 3D information
of the environment centered at the vehicle, and find the ”deepest point” in the scene, which is presumed to be the hallway or an opening, and can be utilized to explore the build. Different algorithms of 3D scene reconstruction are compared in this thesis, and an algorithm to obtain the deepest point in space is developed. Numerical simulations and experiments are demonstrated to verify the feasibility of our algorithm. Results in this these is potentially extendable to the indoor applications of UAVs.

並列關鍵字

Stereo vision ； image processin ； SSD algorithm ； feature point correction ； image correction of the geometric center control

參考文獻

[5] Srigrarom, Sutthiphong, “Development of 3D Feature Detection and On Board Mapping Algorithm from Video Camera for Navigation”, Journal of Applied Science and Engineering, 23-29, 2016.

[9] 陳正霖, 應用立體視覺獲取物體姿態, 淡江大學碩士論文, 新北市, 2009.

[1] Jakob Engel, “Semi-Dense Visual Odometry for a monocular Camera”, IEEE International Conference on Computer Vision, 1449-1456, December 2013.

[2] Larry Matthies, Richard Szeliski, and Takeo Kanade, “Incremental Estimation of Dense Depth Maps from Image Sequences”, IEEE Computer Vision and Pattern Recognition, 366-374, June 1988.

[3] Larry Matthies, Richard Szeliski, “kalman filter-based algorithms for stimating depth from image sequences”, International Conference on Computer Vision,209–236, September 1989.