Prepoznavanje pozicije tijela putem slike i računalnog vida

master's thesis

• Title (english): Computer Vision for Human Pose Estimation
• Author: Mateo Kutnjak
• Mentor: Vladimir Čeperić (mentor)
• Committee member: Vladimir Čeperić (predsjednik povjerenstva)
• Committee member: Slobodan Ribarić (član povjerenstva)
• Committee member: Goran Delač (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2019-07-19, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Problem predikcije poze čovjeka uz dostupne podatke i infrastrukturu postao je rješiv modelima dubokog učenja. Model pješčanog sata sadrži povezane module koji komprimiraju i dekomprimiraju sliku rezidualnim blokovima i zasebnim slojevima konvolucije, maxpool-a i interpolacije te evaluiraju gubitak nekoliko puta kroz mrežu. Rezultantne mape za sve ključne točke nakon svake dekompresije daju vjerojatnost da se na danoj poziciji nalazi ključna točka. Istrenirana mreža prepoznaje poziciju ključne točke na čovjeku iz konteksta iako točka možda nije vidljiva. Efikasnost pristupa vidljiva je u state-of-the-art metodama za prepoznavanje poze čovjeka koje iscrpno koriste modul pješčanog sata.

Abstract (english)

Human pose estimation problem with available data and infrastructure has becomesolvable by deep learning models. Hourglass model contains connected modules whichperform compression and decompression of image tensors with residual blocks andconvolutional,maxpooland interpolation layers. Evaluation of loss is done severaltimes throughout the network. Outputs of network are heatmaps which consist of pixel-wise probability of keypoint presence. Trained network can recognize keypoints evenwhen they are not visible but their locations are obvious. Efficiency of approach isapparent in state-of-the-art methods which exhaustively use hourglass module.

• Keywords: računalni vid; predikcija poze čovjeka; duboko učenje; model pješčanog sata
• Keywords (english): computer vision; human pose estimation; deep learning; hourglass model
• Language: croatian
• URN:NBN: urn:nbn:hr:168:355651
• Study programme: Title: Computing; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2020-01-07 19:35:40