Detekcija akcija u videu korištenjem dubokog učenja

master's thesis

• Title (english): Action Detection in Video using Deep Learning
• Author: Erik Matošević
• Mentor: Zoran Kalafatić (mentor)
• Committee member: Zoran Kalafatić (predsjednik povjerenstva)
• Committee member: Siniša Šegvić (član povjerenstva)
• Committee member: Tomislav Hrkać (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2020-02-21, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Detekcija akcije u videozapisu složen je i izazovan zadatak s ciljem otkrivanja pojave određenih radnji u nekoj situaciji. Temelji se na korištenju dubokih neuronskih mreža koje uče model na označenim slikama od kojih se sastoji video snimka uzimajući u obzir prostornu i vremensku komponentu. U ovome radu dan je uvod u duboko učenje i neuronske mreže. Obuhvaćen je problem detekcije akcija te je opisana C3D arhitektura temeljena na konceptu 3D konvolucijske neuronske mreže koja se pokazala kao efikasno rješenje za navedeni problem. Razvijen je programski okvir za stvaranje skupa podataka te je implementirana C3D neuronska mreža, trenirana na skupu video isječaka s nogometnih utakmica čiji su rezultati opisani i prikazani.

Abstract (english)

Action detection in a video is a complex and challenging task that aims to detect the occurrences of certain actions in a situation. It is based on the use of deep neural networks that train the model on the labeled images from a video by considering the spatial and temporal component. This work provides an introduction to deep learning and neural networks. The action detection problem is covered and a C3D architecture based on the concept of a 3D convolutional neural network is described, which is proven to be an effective solution to this problem. A software framework for dataset creation was developed alongside the C3D neural network implementation. The neural network was trained on a set of video clips from football matches, and the obtained results are described and presented.

• Keywords: računalni vid; detekcija akcije; duboko učenje; konvolucijske neuronske mreže; 3D konvolucijske neuronske mreže; C3D arhitektura
• Keywords (english): computer vision; action detection; deep learning; convolutional neural networks; 3D convolutional neural networks; C3D architecture
• Language: croatian
• URN:NBN: urn:nbn:hr:168:942924
• 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 09:16:58