Detekcija objekata za analizu sportskog videa

undergraduate thesis

• Title (english): Object detection for sport video analysis
• Author: Dino Babić
• Mentor: Zoran Kalafatić (mentor)
• Committee member: Zoran Kalafatić (predsjednik povjerenstva)
• Committee member: Zorana Doždor (član povjerenstva)
• Committee member: Ivan Sabolić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2024-07-03, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Područje računalnog vida jedno je od temeljnih dijelova umjetne inteligencije, s ciljem
prijenosa sposobnosti ljudskog vida na računala. Cilj ovog rada bio je razviti sustav
koji će moći detektirati objekte u sportskim videozapisima, s naglaskom na videozapise
košarkaških utakmica.
U radu je dan pregled razvoja računalnog vida te osnovnih zadataka kao što su klasifikacija
i detekcija objekata. Detaljno su opisane arhitekture R-CNN i njezine nadogradnje,
Fast R-CNN i Faster R-CNN, kao i YOLO arhitektura. Opisane su i mjere uspješnosti
kojima možemo evaluirati rad treniranih modela.
U programskom jeziku Python izgrađena je implementacija R-CNN modela za detekciju
igrača, obruča i lopte u slici te su na tom primjeru još jednom objašnjeni nedostatci
te arhitekture. Opisan je i sustav koji kao ulaz prima videozapis košarkaške utakmice,
a na izlazu daje dva videozapisa. Prvi videozapis predstavlja ulazni videozapis s označenim
omeđujućim pravokutnicima oko igrača, obruča i lopte te omeđujućim mnogokutom
oko terena. Drugi videozapis predstavlja presliku pozicija igrača, koji se nalaze
na terenu, na mapu košarkaškog terena. Sustav je izgrađen koristeći programski jezik
Python. Za detekciju objekata treniran je YOLO model, dok je za segmentaciju terena
korištena biblioteka Detectron2.

Abstract (english)

The field of computer vision is one of the fundamental components of artificial intelligence,
aiming to transfer the capabilities of human vision to computers. The goal of
this paper was to develop a system capable of detecting objects in sports videos, with an
emphasis on basketball game videos.
The paper provides an overview of the development of computer vision and its basic tasks
such as object classification and detection. Detailed descriptions are given of the R-CNN
architecture and its enhancements, Fast R-CNN and Faster R-CNN, as well as the YOLO
architecture. Performance measures for evaluating trained models are also described.
An implementation of the R-CNN model for detecting players, hoops, and balls in images
was developed in the Python programming language, and the shortcomings of this
architecture were further explained using this example. A system which takes a basketball
game video as input and outputs two videos is also described. The first video
represents the input video with bounding boxes around players, hoops, and balls, as well
as a bounding polygon around the court. The second video represents a mapping of
player positions on the court to a basketball court map. The system was developed using
the Python programming language. The YOLO model was trained for object detection,
while the Detectron2 library was used for court segmentation.

• Keywords: računalni vid; detekcija objekata; duboko učenje; strojno učenje; umjetne neuronske mreže
• Keywords (english): computer vision; object detection; deep learning; machine learning; artificial neural networks
• Language: croatian
• URN:NBN: urn:nbn:hr:168:881522
• Study programme: Title: Computing; Study programme type: university; Study level: undergraduate; Academic / professional title: sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva (sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2024-10-12 19:59:59