Multi-target Tracking and Video Synchronization -- {PhD} Thesis Proposal
Identifikátory výsledku
Kód výsledku v IS VaVaI
<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F68407700%3A21230%2F17%3A00308659" target="_blank" >RIV/68407700:21230/17:00308659 - isvavai.cz</a>
Výsledek na webu
<a href="http://cmp.felk.cvut.cz/pub/cmp/articles/smid/Smid-TR-2017-02.pdf" target="_blank" >http://cmp.felk.cvut.cz/pub/cmp/articles/smid/Smid-TR-2017-02.pdf</a>
DOI - Digital Object Identifier
—
Alternativní jazyky
Jazyk výsledku
angličtina
Název v původním jazyce
Multi-target Tracking and Video Synchronization -- {PhD} Thesis Proposal
Popis výsledku v původním jazyce
Multi-target tracking is an active research area with applications in autonomous driving, robotics, life sciences, sports and visual surveillance. Some applications such as sports player tracking and surveillance often employ multiple cameras to deal with occlusion and to improve area coverage. In these cases, precise video synchronization is crucial for further processing. The general aim of the thesis is improvement of multi-target tracking. Lead by practical challenges in an ice hockey dataset we also entered the realm of video synchronization. We build a multi-view multi-target tracking pipeline based on work of Fleuret et al., PAMI 2008; Berclaz et al., PAMI 2011, and generalised possible positioning of cameras to full half-sphere. The multi-view information fusion is based on background subtracted binary image masks. We contributed to this stage by reducing false positive detections when an object is learned as a part of a background model. We present a method for sub-millisecond accurate synchronization of an arbitrary number of cameras using global lighting changes, e.g. photographic flashes. For the sub-millisecond accuracy, the cameras need to be equipped with a rolling shutter image sensor, otherwise the synchronization accuracy is up to whole frames. Large part of our work was published under an open source license.
Název v anglickém jazyce
Multi-target Tracking and Video Synchronization -- {PhD} Thesis Proposal
Popis výsledku anglicky
Multi-target tracking is an active research area with applications in autonomous driving, robotics, life sciences, sports and visual surveillance. Some applications such as sports player tracking and surveillance often employ multiple cameras to deal with occlusion and to improve area coverage. In these cases, precise video synchronization is crucial for further processing. The general aim of the thesis is improvement of multi-target tracking. Lead by practical challenges in an ice hockey dataset we also entered the realm of video synchronization. We build a multi-view multi-target tracking pipeline based on work of Fleuret et al., PAMI 2008; Berclaz et al., PAMI 2011, and generalised possible positioning of cameras to full half-sphere. The multi-view information fusion is based on background subtracted binary image masks. We contributed to this stage by reducing false positive detections when an object is learned as a part of a background model. We present a method for sub-millisecond accurate synchronization of an arbitrary number of cameras using global lighting changes, e.g. photographic flashes. For the sub-millisecond accuracy, the cameras need to be equipped with a rolling shutter image sensor, otherwise the synchronization accuracy is up to whole frames. Large part of our work was published under an open source license.
Klasifikace
Druh
O - Ostatní výsledky
CEP obor
—
OECD FORD obor
10201 - Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)
Návaznosti výsledku
Projekt
—
Návaznosti
S - Specificky vyzkum na vysokych skolach
Ostatní
Rok uplatnění
2017
Kód důvěrnosti údajů
S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů