A study of the effect of noise and occlusion on the accuracy of convolutional neural networks applied to 3D object recognition

Please use this identifier to cite or link to this item: http://hdl.handle.net/10045/72633
Información del item - Informació de l'item - Item information
Title: A study of the effect of noise and occlusion on the accuracy of convolutional neural networks applied to 3D object recognition
Authors: Garcia-Garcia, Alberto | Garcia-Rodriguez, Jose | Orts-Escolano, Sergio | Oprea, Sergiu | Gomez-Donoso, Francisco | Cazorla, Miguel
Research Group/s: Informática Industrial y Redes de Computadores | Robótica y Visión Tridimensional (RoViT)
Center, Department or Service: Universidad de Alicante. Departamento de Tecnología Informática y Computación | Universidad de Alicante. Departamento de Ciencia de la Computación e Inteligencia Artificial | Universidad de Alicante. Instituto Universitario de Investigación Informática
Keywords: Deep learning | 3D object recognition | Convolutional neural networks | Noise | Occlusion | Caffe
Knowledge Area: Arquitectura y Tecnología de Computadores | Ciencia de la Computación e Inteligencia Artificial
Issue Date: Nov-2017
Publisher: Elsevier
Citation: Computer Vision and Image Understanding. 2017, 164: 124-134. doi:10.1016/j.cviu.2017.06.006
Abstract: In this work, we carry out a study of the effect of adverse conditions, which characterize real-world scenes, on the accuracy of a Convolutional Neural Network applied to 3D object class recognition. Firstly, we discuss possible ways of representing 3D data to feed the network. In addition, we propose a set of representations to be tested. Those representations consist of a grid-like structure (fixed and adaptive) and a measure for the occupancy of each cell of the grid (binary and normalized point density). After that, we propose and implement a Convolutional Neural Network for 3D object recognition using Caffe. At last, we carry out an in-depth study of the performance of the network over a 3D CAD model dataset, the Princeton ModelNet project, synthetically simulating occlusions and noise models featured by common RGB-D sensors. The results show that the volumetric representations for 3D data play a key role on the recognition process and Convolutional Neural Network can be considerably robust to noise and occlusions if a proper representation is chosen.
Sponsor: This work has been supported by the Spanish Government DPI2013-40534-R grant for the SIRMAVED project, also supported with FEDER funds. This work has also been funded by the grant “Ayudas para Estudios de Máster e Iniciación a la Investigación” from the University of Alicante.
URI: http://hdl.handle.net/10045/72633
ISSN: 1077-3142 (Print) | 1090-235X (Online)
DOI: 10.1016/j.cviu.2017.06.006
Language: eng
Type: info:eu-repo/semantics/article
Rights: © 2017 Elsevier Inc.
Peer Review: si
Publisher version: http://dx.doi.org/10.1016/j.cviu.2017.06.006
Appears in Collections:INV - I2RC - Artículos de Revistas
INV - RoViT - Artículos de Revistas

Files in This Item:
Files in This Item:
File Description SizeFormat 
Thumbnail2017_Garcia-Garcia_etal_CompVisImageUnderst_final.pdfVersión final (acceso restringido)2,74 MBAdobe PDFOpen    Request a copy
Thumbnail2017_Garcia-Garcia_etal_CompVisImageUnderst_preprint.pdfPreprint (acceso abierto)3,62 MBAdobe PDFOpen Preview


Items in RUA are protected by copyright, with all rights reserved, unless otherwise indicated.