Motion-Based Tracking of Multiple Low-Relative-Depth Objects

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F68407700%3A21230%2F12%3A00200616" target="_blank" >RIV/68407700:21230/12:00200616 - isvavai.cz</a>

Výsledek na webu

—

DOI - Digital Object Identifier

—

Jazyk výsledku

angličtina

Název v původním jazyce

Motion-Based Tracking of Multiple Low-Relative-Depth Objects

Popis výsledku v původním jazyce

In this report, we present a method for online tracking of multiple rigid low-relative-depth objects or object surfaces from a single moving or static camera. The method determines the number of tracked entities in the scene automatically and initialisesthem without any human interaction or by running a specific object detector. Tracking is formulated as an energy-based multi-model fitting problem based on displacement of a semi-dense set of local Lukas-Kanade trackers. An over-complete set of motion models is generated in each frame and fitted considering the motion similarity. For each detected surface the motion model is propagated in time and to ensure temporal consistency and locality of the tracked entities in the scene two novel energy terms are introduced. We demonstrate the abilities of the proposed algorithm on several long real-world sequences. The algorithm works at 2-6fps depending on the complexity of the scene.

Název v anglickém jazyce

Motion-Based Tracking of Multiple Low-Relative-Depth Objects

Popis výsledku anglicky

In this report, we present a method for online tracking of multiple rigid low-relative-depth objects or object surfaces from a single moving or static camera. The method determines the number of tracked entities in the scene automatically and initialisesthem without any human interaction or by running a specific object detector. Tracking is formulated as an energy-based multi-model fitting problem based on displacement of a semi-dense set of local Lukas-Kanade trackers. An over-complete set of motion models is generated in each frame and fitted considering the motion similarity. For each detected surface the motion model is propagated in time and to ensure temporal consistency and locality of the tracked entities in the scene two novel energy terms are introduced. We demonstrate the abilities of the proposed algorithm on several long real-world sequences. The algorithm works at 2-6fps depending on the complexity of the scene.

Druh

O - Ostatní výsledky

CEP obor

JD - Využití počítačů, robotika a její aplikace

OECD FORD obor

—

Projekt

<a href="/cs/project/GBP103%2F12%2FG084" target="_blank" >GBP103/12/G084: Centrum pro multi-modální interpretaci dat velkého rozsahu</a><br>

Návaznosti

P - Projekt vyzkumu a vyvoje financovany z verejnych zdroju (s odkazem do CEP)

Rok uplatnění

2012

Kód důvěrnosti údajů

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů