Please use this identifier to cite or link to this item:
Title: Detecció visual de bucles en fons marins mitjançant xarxes neuronals
Author: Gálvez Santos, Jordi
Director: Antonio Burguera Burguera
Tutor: Carles Ventura Royo
Keywords: SLAM, Computer vision, Deep Learning
Issue Date: 2-Jun-2020
Publisher: Universitat Oberta de Catalunya (UOC)
Abstract: An autonomous vehicle placed in a not known environment must be able to be located and moved without supervision. SLAM wants to resolve this problem. Identify when a robot has previously passed through a given point is called loop closure and is an important step on SLAM methodology. There are various mechanisms for resolve loop closure problem. Compare images with machine learning is one of these techniques. The purpose of this work is to create a Neural Network to detect loop closures on seabed. The loop closure detection is performed by images of seabed taken by robots. The input images in Neural Network will be transformed to create a synthetic image, generating a pair of images (the original and the altered). A global descriptor (HOG) will be calculated on one of these loop images and the Network will be trained with the other image to find a similar descriptor. With this training, the Neural Network will learn to generate similar descriptors from pairs of images involved in loop closures On the resolution of this problem we will use Keras on Python. We built a Neural Network that has been evaluated and then refined to provide better results.
Language: Catalan
Appears in Collections:Bachelor thesis, research projects, etc.

Files in This Item:
File Description SizeFormat 
jordigsTFM0620.pdfMemòria del TFM17.08 MBAdobe PDFView/Open
jordigsPresentacio.mp4Presentació del TFM145.4 MBMP4View/Open
jordigsCode.zipImplementació principal de la xarxa neuronal21.05 kBUnknownView/Open

This item is licensed under a Creative Commons License Creative Commons