Detekcija i praćenje igrača u snimkama nogometnih utakmica

master's thesis

• Title (english): Player detection and tracking in football match video
• Author: Andi Škrgat
• Mentor: Zoran Kalafatić (mentor)
• Committee member: Zoran Kalafatić (predsjednik povjerenstva)
• Committee member: Tomislav Hrkać (član povjerenstva)
• Committee member: Vladimir Čeperić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2023-07-03, Croatia

Abstract

Analiza nogometnih utakmica zasnovana na matematičkim alatima postaje sve popularnija. Ovaj rad pokušava odgovoriti na pitanje mogu li se moderne tehnike detekcije i praćenja objekata primijeniti za izvlačenje informacija korisnih analitičarima, trenerima i skautima iz snimaka nogometnih utakmica. Eksperimenti pokazuju kako detekcija igrača radi odlično dok je detekcija lopte nešto lošija, ali je i sam izazov puno teži. Jedan od glavnih razloga možemo pronaći u poprilično velikoj udaljenosti s koje je utakmica snimana i s koje je lopta na slici jako mala. Iako pokazuje odlične rezultate na MOT metrikama, HOTA metrika pokazuje kako model praćenja nema dovoljno kapaciteta za praćenje svih igrača kroz čitavu utakmicu. Problemu se doskočilo na način da je razvijena aplikacija pomoću koje korisnik može voditi interakciju sa sustavom i između ostaloga, ispravljati njegove nedostatke. Konačna implementacija pokazuje kako je moguće dobiti značajnu količinu statističkih informacija korisnih za nogometne profesionalce.

Abstract (english)

Football match analysis based on mathematical tools is becoming more and more popular. The thesis tries to answer the question of how applicable state-of-the-art object detection and tracking techniques are for extracting information from match recordings that can be used for football analysts, coaches and scouts. The conducted experiments clearly show object detection works almost perfectly with players and sufficiently well for the ball. The ball is not as perfectly tracked as the person and of the reasons can be found in quite the large distance from which the camera records and from which the ball appears very small. Although it shows great results on multiple object tracking (MOT) metrics, the HOTA tracking metric reveals the tracker's problem of continuously assigning correct identifiers to the objects. With an application that serves as an interface for human communication, the thesis shows the presented implementation can be used by football analysts to extract a variety of statistical information.

• Keywords: računalni vid; detekcija objekata; praćenje objekata; YOLOv5; duboko učenje; nogomet; model kontrole terena
• Keywords (english): computer vision; object detection; object tracking; YOLOv5; deep learning; football; pitch control model
• Language: english
• URN:NBN: urn:nbn:hr:168:542964
• 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: 2023-10-11 10:40:05