Automatic Animal Species Identification Based on Camera Trapping DataReportar como inadecuado

Automatic Animal Species Identification Based on Camera Trapping Data - Descarga este documento en PDF. Documentación en PDF para descargar gratis. Disponible también para leer online.

Animal Species Identification, Support Vector Machines, Average Pooling, Fisher Vector, Camera Trapping, Gaussian Mixture Model

Wang, Baoliang

Supervisor and department: Zaiane, Osmar Computing Science

Examining committee member and department: Bayne, Erin Biological Sciences Boulanger, Pierre Computing Science

Department: Department of Computing Science


Date accepted: 2014-08-26T11:18:52Z

Graduation date: 2014-11

Degree: Master of Science

Degree level: Master's

Abstract: The classification of animal images based on camera trapping data is an importantand challenging task in the domains of computer vision, machine learning and ecological management. This thesis presents an animal species identification system that can automatically identify the species of an animal captured in an image by a camera trap. We use the Fisher Vector coding approach on top of the dense Scale Invariant Feature Transform features and the cell-structured Local Binary Pattern descriptors to generate a fixed length vector representation for each image and then feed this vector representation to the linear Support Vector Machines for learning and classification. Unlike traditional Bag of Visual Words models that only use a generative method or discriminative method, the powerful Fisher Kernel framework combines the advantages of both generative and discriminative approaches to encode image descriptors and then classify images. The key idea is to characterize an image with a gradient vector derived from a generative probability model and to subsequently feed this gradient vector to a discriminative classifier. Instead of only using zero-order image statistics like in conventional approaches, the Fisher Vector coding method retains zero-order, first-order and second-order information and thus allows less image approximation error. Extensive experimental study shows that our method achieves the highest classification accuracy compared to various conventional.

Language: English

DOI: doi:10.7939-R3V97ZZ5X

Rights: Permission is hereby granted to the University of Alberta Libraries to reproduce single copies of this thesis and to lend or sell such copies for private, scholarly or scientific research purposes only. Where the thesis is converted to, or otherwise made available in digital form, the University of Alberta will advise potential users of the thesis of these terms. The author reserves all other publication and other rights in association with the copyright in the thesis and, except as herein before provided, neither the thesis nor any substantial portion thereof may be printed or otherwise reproduced in any material form whatsoever without the author's prior written permission.

Autor: Wang, Baoliang



Documentos relacionados