3D indoor reconstruction using depth-map-based scene complexity analysis guided kinectfusion

Name: Somkiat Khamphuea

keyword: 3D reconstruction

; KinectFusion

; Surface normal

; Scene complexity

Abstract: This thesis presents a novel approach for 3D reconstruction based on KinectFusion. The iterative closest point algorithm (ICP) employed in KinectFusion works well when there are sufficient 3D features in a scene to be reconstructed. Conversely, it is difficult to reconstruct simple scenes with limited 3D features such as planar structures. We propose to use visual odometry (VO), in place of ICP, when only insufficient 3D features are available in a scene. Regardless of whether there are sufficient 3D features or not, VO works well as long as the scene contains sufficient 3D features such as textures and corner points. The proposed method then automatically selects ICP or VO, depending on the complexity of the scene. The complexity of the scene is evaluated with the magnitudes of the discontinuities in surface normal vectors in depth maps. Experimental results show that the proposed method outperforms the methods based on either ICP or VO alone

Thammasat University. Thammasat University Library

Address: BANGKOK

Email: preserv@tu.ac.th

Name: Kondo, Toshiaki

Role: advisor

Name: Itthisek Nilkhamhang

Role: co-advisor

Created: 2015

Modified: 2021-10-19

Issued: 2021-10-19

วิทยานิพนธ์/Thesis

application/pdf

DOI: 10.14457/TU.the.2015.21

eng

DegreeName: Master of Engineering

Level: Master's Degree

Descipline: Information and Communication Technology for Embedded Systems

Grantor: Thammasat University

©copyrights Thammasat University

ลำดับที่.	ชื่อแฟ้มข้อมูล	ขนาดแฟ้มข้อมูล	จำนวนเข้าถึง	วัน-เวลาเข้าถึงล่าสุด
1	4685somkiat.pdf	2.11 MB

ใช้เวลา

0.028589 วินาที