Afektivno računarstvo na ugradbenim računalima

master's thesis

• Title (english): Affective Computing on Embedded Platforms
• Author: Denis Malić
• Mentor: Davor Petrinović (mentor)
• Committee member: Davor Petrinović (predsjednik povjerenstva)
• Committee member: Marko Subašić (član povjerenstva)
• Committee member: Tomislav Petković (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2018-07-11, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Electrical Engineering

Abstract

Korištenje emocija u interakciji čovjeka sa računalom je povezano sa praktičnim aplikacijama koje su postale svakodnevnica. Unatoč tome, izazovi koje sustavi koji imaju cilj klasificirati emocionalna stanja predstavljaju su veliki. Počevši od neodređenosti emocionalnih granica do računske zahtjevnosti često upotrebljavanih algoritama i funkcija koje implementiraju klasifikatore. U ovom projektu je korišteno ugradbneno računalo za potrebe prepoznavanja afektivnih stanja koje uključuju: računanje značajki i statističkih mjera, prilagodbu baze, treniranje klasifikatora, evaluaciju točnosti i mjerenje vremena izvođenja implementiranih algoritama. Zadaće ovog projekta su izvršene korištenjem openSMILE alata za računanje značajki i Python funkcija za implementiranje klasifikatora. Nakon što je procijenjena upotrebljivost više klasifikatora uz razne parametre, SVM klasifikator je treniran sa skupom za treniranje i komparativna evaluacija u vidu vremena izvođenja i točnosti je izvršena. Točnost je usporediva sa ostalim radovima. Vrijeme izvođenja na Raspberry Pi pokazuje poboljšanja sa smanjenjem frekvencije uzorkovanja, ali ne zadovoljava marginu za izvođenje u stvarnom vremenu. Zadovoljavajuća točnost je djelom posljedica pažljive prilagodbe baze koja osigurava da su afektivna stanja približno jednako zastupljena u skupu za treniranje i testiranje.

Abstract (english)

Utilising emotion information through human-computer interactions is associated with practical applications that are beginning to pervade our every-day life. However, the challenges systems which aim to classify emotive states pose are many. Starting from the fuzziness of emotional boundaries to the computational demand of widely used algorithms and classifier functions. In this project an embedded platform was used for an affect recognition task which included computing features and functionals, database preparation, fitting data to a classifier, an evaluation of accuracy and timing the embedded board implemented algorithms. Since computation time is a critical factor in embedded platforms’ performance evaluation, the importance of timing the algorithms was emphasised. The tasks of this project were facilitated by making use of the openSMILE toolkit and Python functions for computing the features and implementing the classifier respectively. Following the assessment of different classifiers under diverse parametrisation, the prepared training data set was fitted to the SVM classifier and prediction evaluation was performed in terms of accuracy and execution time, both of which were compared with embedded platform’s results. Accuracy obtained is comparable to other works. The running time on the Raspberry Pi shows improvement as we decrease the sampling rate, but it is above the acceptable margin for real-time applications. The high accuracy is partially indebted to the meticulous database preparation, which ensured that the two data sets were evenly filled with emotive content from different octants.

• Keywords: strojno učenje; Python; SVM; prepoznavanje emocija; afektivno računarstvo; afekt; emocionalno stanje; Raspberry Pi
• Keywords (english): machine learning; Python; SVM; emotion recognition; affective computing; affect; emotive state; Raspberry Pi
• Language: english
• URN:NBN: urn:nbn:hr:168:801443
• 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: 2019-03-13 22:32:02