Analiza videozapisa nogometnih utakmica tehnikama računalnog vida

master's thesis

• Title (english): Analysis of soccer match video recordings using computer vision techniques
• Author: Tihomir Pavić
• Mentor: Tomislav Hrkać (mentor)
• Committee member: Tomislav Hrkać (predsjednik povjerenstva)
• Committee member: Siniša Šegvić (član povjerenstva)
• Committee member: Zoran Kalafatić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2024-07-02, Croatia

Abstract

Cilj ovog rada bio je razviti i ispitati metodu koja može unutar videozapisa cijele
nogometne utakmice detektirati ključne nogometne akcije s ciljem kreiranja video sažetaka
nogometne utakmice. Metoda se sastoji od dvije konvolucijske neuronske mreže. Prvi dio
čini arhitektura ResNet-152 koja iz ulaznog slijeda okvira (engl. frames) na izlazu daje
slijed izdvojenih značajki. Sekvence tih značajki ulaze u drugi model imena RMS-Net koji
za svaku sekvencu mora odrediti nalazi li se neka od ključnih nogometnih akcija unutar te
sekvence radeći klasifikaciju sekvence u jednu od sedam klasa te za njih i predvidjeti
relativni vremenski pomak kako bi znali trenutak u kojem se akcija dogodila. Ako model
ne pronađe akciju unutar sekvence, tada tu sekvencu klasificira kao klasu pozadine koja ne
ulazi u konačni video sažetak.

Abstract (english)

The aim of this thesis was to develop and test a method that can detect key soccer actions
within a full match video to create video summaries of the soccer match. The method
consists of two convolutional neural networks. The first part is a ResNet-152 architecture,
which extracts a sequence of features from the input sequence of frames. The sequences of
these features are fed into a second model named RMS-Net, which for each sequence must
determine whether any of the key soccer actions are present within that sequence by
classifying the sequence into one of seven classes and predicting the relative temporal
offset to know the exact moment the action occurred. If the model does not find an action
within the sequence, it classifies that sequence as a background class, which is not included
in the final video summary.

• Keywords: videozapis; konvolucijske neuronske mreže; ResNet-152; detektiranje nogometnih akcija
• Keywords (english): video; convolutional neural networks; ResNet-152; action spotting in football
• Language: croatian
• URN:NBN: urn:nbn:hr:168:333947
• 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: 2024-10-12 21:47:53