Please use this identifier to cite or link to this item:
|Title:||Robust 3D reconstruction in adverse condition||Authors:||Xiao, Renxiang||Keywords:||Engineering::Electrical and electronic engineering||Issue Date:||2022||Publisher:||Nanyang Technological University||Source:||Xiao, R. (2022). Robust 3D reconstruction in adverse condition. Master's thesis, Nanyang Technological University, Singapore. https://hdl.handle.net/10356/162541||Abstract:||3D reconstruction refers to the mathematical process and computer technology of recovering 3D information of an object using 2D projection, including data acquisition, preprocessing, point cloud construction and stitching. Image-based 3D reconstruction is a method of extracting 3D information of a scene from multiple pictures and reconstructing a 3D model of the scene. 3D reconstruction technology has a wide range of applications in autonomous driving, virtual reality, smart home, cultural relic reconstruction and other fields. But most reconstructions are done under normal conditions, while 3D reconstructions fail under adverse conditions like smoke or fog. The purpose of this paper is to perform color point cloud reconstruction under adverse conditions. Aiming at the problem of data acquisition under unfavorable conditions, a new suite is designed as a data acquisition platform, and human interference is generated during the data acquisition process. To demonstrate how reconstruction under adverse conditions is affected, a completed 3D reconstruction algorithms are run on the same dataset as the ground truth. Aiming at the problem that front-end odometer calculation cannot be realized under adverse conditions, a new method based on RADAR-Thermal reconstruction is proposed. Use Loftr to obtain feature points from thermal images, map radar points to thermal images, and obtain depth information of feature points. By matching each point cloud submap, we can obtain a global map of the scene. Experimental results obtained using ground truth after running the algorithm on the collected dataset can demonstrate that the reconstruction method we developed still maintains a certain accuracy and robustness under adverse conditions.||URI:||https://hdl.handle.net/10356/162541||Fulltext Permission:||restricted||Fulltext Availability:||With Fulltext|
|Appears in Collections:||EEE Theses|
Updated on Nov 30, 2022
Updated on Nov 30, 2022
Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.