Sports Tracking by Correlating Thermal Video with Inertia Data
Authors
Bodin, Niels Dyremose ; Røntved, Mike
Term
4. term
Publication year
2017
Submitted on
2017-06-24
Pages
94
Abstract
Dette projekt undersøger, hvordan termisk video (som viser varme fremfor farver) kan kobles med data fra accelerometre (sensorer, der måler bevægelse og acceleration) for at identificere, hvilken måling hører til hvilken person. Vi udviklede en behandlingskæde, som kan opdage spillere i termiske optagelser, følge dem over tid og danne korte spor (tracklets). Ud fra sporene estimerer systemet acceleration i videoen og sammenligner den med accelerationen fra accelerometrene for at matche hver sensor til de rigtige spor. Rapporten gennemgår alle trin—detektion, tracking og matching—som alle er designet, implementeret og testet. Som proof-of-concept blev systemet afprøvet på et datasæt fra en halv fodboldbane i Gigantium sports arena, hvor deltagerne spillede fodbold. Resultaterne er lovende og viser, at termisk video kan korreleres med accelerometre, men trackingen kræver mere arbejde, før systemet kan indgå i et automatisk setup.
This project explores how to link thermal video (which shows heat rather than color) with accelerometer data (sensors that measure movement and acceleration) to identify which measurements belong to which person. We developed a processing pipeline that detects players in thermal footage, tracks them over time to produce short tracks (tracklets), estimates acceleration from those tracks, and compares it with acceleration from accelerometers to match each sensor to the correct track. The report covers all stages—detection, tracking, and matching—which were designed, implemented, and tested. As a proof of concept, the system was evaluated on a dataset recorded on half a football field in the Gigantium sports arena while participants played football. The results are promising and show that thermal video can be correlated with accelerometer data, but the tracking component needs further work before the system can be used in an automatic setup.
[This abstract was generated with the help of AI]
Keywords
computer vision ; tracking ; detection ; thermal ; correlation ; accelerometer ; occlusion
Documents