Prepoznavanje emocija pomoću robotskog vida i dubokog učenja

master's thesis

• Title (english): Recognition of Emotions using Robot Vision and Deep Learning
• Author: Zvonimir Jukić
• Mentor: Zdenko Kovačić (mentor)
• Committee member: Zdenko Kovačić (predsjednik povjerenstva)
• Committee member: Stjepan Bogdan (član povjerenstva)
• Committee member: Matko Orsag (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2019-07-18, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Electrical Engineering

Abstract

Ljudi, kao socijalna bića, ovise o komunikaciji s drugim ljudima. Većina komunikacije sačinjena je od neverbalnih komponenti, a emocije, i izraz lica općenito, nose veliki dio te informacije. Prepoznavanje emocija je bitan alat za kvalitetniju interakciju ljudi i robota, a upotrebu pronalazi i u drugim područjima. Mnogi visoko stresni poslovi zahtijevaju dodatne mjere opreza na radnom mjestu, a i pri selekciji, što potiče na istraživanje korisnosti informacije o emociji u svrhu prepoznavanja kompleksnijih stanja poput umora, stresa ili kognitivnog preopterećenja. U svrhu razvijanja alata za prepoznavanje emocija pomoću robotskog vida i dubokog učenja detaljno je proučena literatura po pitanju prepoznavanja emocija, baza podataka i dubokog učenja. Opisani su različiti pristupi klasifikaciji emocija i različiti modeli opisa emocija u kontekstu izraza lica. Spomenute su javne baze podataka koje se najčešće koriste u svrhu istraživanja u području prepoznavanja izraza lica, s naglaskom na bazu CK+, koja je korištena za trening i evaluaciju mreže. Najviše pažnje posvećeno je dubokom učenju, s naglaskom na konvolucijske neuronske mreže koje su korištene kao temelj za izradu ovog alata. Navedeni su svi modeli koji su trenirani, zajedno s rezultatima i svim relevantnim mjerama vrednovanja modela. Rezultati modela uspoređeni su s rezultatima srodnih istraživanja na već spomenutim bazama podataka. Unatoč oskudnom setu podataka, postignuti su rezultati sumjerljivi najboljim rezultatima srodnih istraživanja.

Abstract (english)

People, as social beings, depend on communicating with other people. Most of the communication is made up of non-verbal components, and emotions, as well as facial expressions in general, carry a large part of that information. Emotion recognition is an important tool for increasing the quality of human-robot interaction, but it also finds usage in other areas. Many highly stressful jobs require additional precautions in the workplace, as well as during selection, encouraging research into the usefulness of information on emotions in recognizing more complex states such as fatigue, stress or cognitive overload. For the purpose of developing an emotion recognition tool using robotic vision and deep learning, literature on emotion, databases and deep learning has been studied in detail. Different approaches to the classification of emotions and different models of describing emotions in the context of facial expressions have been described. Public databases which are most commonly used in the research field of facial expression recognition have been mentioned, with a highlight on the CK+ database which has been used for the training and evaluation of the network. Most attention has been devoted to deep learning, with an emphasis on convolutional neural networks that have been used as the basis for creating this tool. All models which have been trained were listed along with the results and all relevant measures of model evaluation. The results of the models have been compared with the results of related research on the above-mentioned databases. Despite a scarce set of data, results which measure up to the best results in related research have been achieved.

• Keywords: Emocije; prepoznavanje emocija; baza podataka; duboko učenje; neuronske mreže; konvolucijske neuronske mreže; strojno učenje; robotski vid
• Keywords (english): Emotion; emotion recognition; database; deep learning; neural networks; convolutional neural networks; machine learning; robot vision
• Language: croatian
• URN:NBN: urn:nbn:hr:168:064519
• Study programme: Title: Electrical Engineering and Information Technology; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka elektrotehnike i informacijske tehnologije (magistar/magistra inženjer/inženjerka elektrotehnike i informacijske tehnologije)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2020-01-07 18:47:08