Računalni modeli za procjenu antropometrijskih mjera čovjeka iz slika

master's thesis

• Title (english): Estimating Anthropometric Measurements from Images
• Author: Mateo Puhalović
• Mentor: Tomislav Pribanić (mentor)
• Committee member: Tomislav Pribanić (predsjednik povjerenstva)
• Committee member: Marko Subašić (član povjerenstva)
• Committee member: Hrvoje Džapo (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2021-07-06, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

U ovome radu opisan je problem procjene antropometrijskih mjera čovjeka iz slika te su opisana 2 fundamentalno različita pristupa: tradicionalni i pristup dubokim učenjem. Tradicionalni pristup se bazira na pronalasku ključnih točaka i siluete osobe te računanjem duljina na silueti korištenjem istih. Sa dobivenim duljinama se zatim aproksimiraju dijelovi tijela jednostavnim oblicima čije se dimenzije smatraju dimenzijama aproksimiranih dijelova tijela. Pristup dubokim učenjem se bazira na predviđaju parametara statističkog 3D modela korištenjem silueta osobe te mjerenjem dimenzija tijela iz tako dobivenog 3D modela. Pristup dubokim učenjem stvara modele za muške i ženske osobe odvojeno. Opisane implementacije su zatim uspoređene na skupu subjekata iz čijih se slika računaju dimenzije tijela.

Abstract (english)

This paper describes the problem of estimating human anthropometric measures from images and describes 2 fundamentally different approaches: traditional and deep learning approach. The traditional approach is based on finding the key points and silhouettes of a person and calculating the lengths on the silhouette using that information. With the obtained lengths, the body parts are then approximated by simple shapes whose dimensions are considered to be the dimensions of the approximated body parts. The deep learning approach is based on predicting the parameters of a statistical 3D model using the silhouette of a person and measuring the dimensions of the body from the 3D model thus obtained. The deep learning approach creates models for men and women separately. The described implementations were then compared on a set of subjects from whose images body dimensions are calculated.

• Keywords: antropometrija; duboko učenje; strojno učenje; računalni vid; Keras; Tensorflow; segmentacija slike; 3D model osobe; ResNet; neuronska mreža
• Keywords (english): anthropometry; deep learning; machine learning; computer vision; Keras; Tensorflow; image segmentation; 3D model of a person; ResNet; neural network
• Language: croatian
• URN:NBN: urn:nbn:hr:168:885212
• 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: 2023-01-13 15:58:09