kth.sePublications KTH
Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Multi-view Mesh Estimation of Football Players
KTH, School of Electrical Engineering and Computer Science (EECS).
2025 (English)Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesisAlternative title
Flervybaserad mesh-estimering av fotbollsspelare (Swedish)
Abstract [en]

Human Mesh Recovery (HMR) has achieved significant progress in controlled settings, but its application to real-world sports scenarios remains largely unexplored. This thesis presents a systematic evaluation of state-of-the-art HMR techniques for reconstructing 3D meshes of football players from production-grade stadium camera footage. Unlike standard HMR datasets, football presents unique challenges including rapid player movements, frequent occlusions, loose clothing, low-resolution imagery, and the absence of ground-truth mesh annotations for supervision. Conducted in collaboration with Tracab, a leading sports analytics company, this research employs domain-specific fine-tuning of an HMR model using real multi-view stadium camera images. The methodology addresses the lack of ground-truth annotations through indirect supervision, requiring careful consideration of alignment issues between different keypoint definitions. Experimental results demonstrate good mesh prediction accuracy on typical player poses, varying lighting conditions, and multiple camera perspectives across different stadiums. However, limitations remain in challenging scenarios such as player occlusions and non-standard poses. The findings suggest that domain-specific adaptation of HMR models holds significant promise for commercial sports applications, while highlighting key areas for future research including temporal consistency, multi-player interactions, and computational optimization. 

Abstract [sv]

Återvinning av mänskliga mesh (HMR) har uppnått betydande framsteg i kontrollerade miljöer, men dess tillämpning på verkliga idrottsscenarion förblir till stor del outforskad. Denna avhandling presenterar en systematisk utvärdering av toppmoderna HMR-tekniker för rekonstruktion av 3D-mesh av fotbollsspelare från professionella stadionkamerabilder. Till skillnad från standardiserade HMR-dataset presenterar fotboll unika utmaningar inklusive snabba spelarrörelser, frekventa ocklusioner, löst sittande kläder, lågupplösta bilder och frånvaron av grund-sanning mesh-annoteringer för övervakning. Genomförd i samarbete med Tracab, ett ledande sportanalysföretag, använder denna forskning domänspecifik finjustering av en HMR-modell med hjälp av verkliga flervinkla stadionkamerabilder. Metodiken adresserar bristen på grund-sanning annoteringer genom indirekt övervakning, vilket kräver noggrann hänsyn till justeringsproblem mellan olika keypointdefinitioner. Experimentella resultat visar god mesh-prediktionsnoggrannhet på typiska spelarposer, varierande ljusförhållanden och flera kameraperspektiv över olika stadioner. Begränsningar kvarstår dock i utmanande scenarier såsom spelarocklusioner och icke-standardposer. Resultaten tyder på att domänspecifik anpassning av HMR-modeller har betydande potential för kommersiella sporttillämpningar, samtidigt som de belyser viktiga områden för framtida forskning inklusive temporal konsistens, flerspelarinteraktioner och beräkningsoptimering.

Place, publisher, year, edition, pages
Stockholm: KTH Royal Institute of Technology , 2025. , p. 59
Series
TRITA-EECS-EX ; 2025:707
Keywords [en]
Human Mesh Recovery, 3D Human Pose Estimation, Football
Keywords [sv]
Mänsklig Mesh-återställning, 3D Mänsklig Posuppskattning, Fotboll
National Category
Computer Sciences Computer Engineering
Identifiers
URN: urn:nbn:se:kth:diva-372538OAI: oai:DiVA.org:kth-372538DiVA, id: diva2:2012482
External cooperation
TRACAB
Educational program
Master of Science - Machine Learning
Supervisors
Examiners
Available from: 2025-11-18 Created: 2025-11-08 Last updated: 2025-11-20Bibliographically approved

Open Access in DiVA

fulltext(13245 kB)312 downloads
File information
File name FULLTEXT01.pdfFile size 13245 kBChecksum SHA-512
6e2debc78b72f496499f9395d72aafb751e6a6721672b0ef6c8787e377f277aa22a0e52c4ee3cf56ada9297986a0d5e1679f1b7614b28c89c13b010628106198
Type fulltextMimetype application/pdf

By organisation
School of Electrical Engineering and Computer Science (EECS)
Computer SciencesComputer Engineering

Search outside of DiVA

GoogleGoogle Scholar
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

urn-nbn

Altmetric score

urn-nbn
Total: 403 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf