Javascript must be enabled for the correct page display

An Autonomous Indoor Navigation System Based on Visual Scene Recognition Using Deep Neural Networks

Bidoia, F. (2017) An Autonomous Indoor Navigation System Based on Visual Scene Recognition Using Deep Neural Networks. Master's Thesis / Essay, Artificial Intelligence.

[img]
Preview
Text
thesis.pdf - Published Version

Download (10MB) | Preview
[img] Text
toestemming.pdf - Other
Restricted to Backend only

Download (77kB)

Abstract

In this thesis we propose a novel approach for Indoor Navigation Systems that is based on visual information. This new method discards the need for precise Global Localization while adopting a more human-like approach. For this aim, we deeply analyze the current state of the art of computer vision, comparing the classical methods such as Scale Invariant Feature Transform (SIFT) and Visual Bag of Words, with the most recent successes of Convolutional Neural Networks in this field. A further analysis of the state of the art of Deep Neural Networks for image classification is proposed, which focuses on the similarities and differences when compared to navigation tasks. Based on this analysis, we developed a novel Deep Neural Network architecture that takes inspiration from the most recent Inception V3 architecture. The results obtained from specifically designed tests show how Visual Navigation tasks rely on geometrical properties of the scene. Although previous deep learning architectures have often made use of techniques such as pooling, our architecture does not use this. In fact, we show how our neural network significantly outperforms the state of the art of image classification in the particular task of visual navigation.

Item Type: Thesis (Master's Thesis / Essay)
Supervisor name: Schomaker, dr. L.R.B. and Shantia, A.
Degree programme: Artificial Intelligence
Thesis type: Master's Thesis / Essay
Language: English
Date Deposited: 15 Feb 2018 08:31
Last Modified: 02 May 2019 09:24
URI: https://fse.studenttheses.ub.rug.nl/id/eprint/15816

Actions (login required)

View Item View Item