Inferring Depth Maps from 2-Dimensional Laser Ranging Data in a Simulated Environment

Loading...
Thumbnail Image

URL

Journal Title

Journal ISSN

Volume Title

Sähkötekniikan korkeakoulu | Master's thesis

Date

2018-12-17

Department

Major/Subject

Control, Robotics and Autonomous Systems

Mcode

ELEC3025

Degree programme

AEE - Master’s Programme in Automation and Electrical Engineering (TS2013)

Language

en

Pages

39+10

Series

Abstract

Depth estimation plays a key role in mobile robotics for applications including scene understanding, navigation and mapping. Recently, deep learning methods have proven effective in estimating depth maps from a combination of different sources such as 3D LiDAR or RGB images. However, they face two challenges; the lack of dense ground truth data and the depth input sparsity, which ranges from 4-10% pixel density on an input image. This thesis explores the feasibility of inferring a full depth map from extremely sparse 2D LiDAR measurements via neural network. To address the lack of ground truth data, a simulation tool is created for data gathering. The results show that from our sparse input of 0.024% pixel density on input images, the tested network infers shapes but struggles with blurry boundaries on objects.

Description

Supervisor

Kyrki, Ville

Thesis advisor

Verdoja, Francesco
Lundell, Jens

Keywords

depth completion, deep learning, robotic simulation, convolutional neural networks

Other note

Citation