Sistem za sledenje znanim objektom v spremenljivih svetlobnih pogojih na sceni s pomočjo posnetkov IR in vidnega spektra

Meden, Blaž (author), Peer, Peter (mentor)

V tem raziskovalnem delu naslovimo problematiko sledenja večih objektov pod vplivom negotovih svetlobnih razmer. Da bi rešili problem dinamične osvetlitve implementiramo dva dobro znana sledilna pristopa, ki operirata v dveh ločenih domenah. Pri tem se prvi sledilnik zanaša na aktivni sistem oznak, ki deluje v bližnjem infrardečem spektru, drugi pa izkorišča najsodobnejši obstoječi detektor oseb in barvno infomacijo opazovane scene v vidnem spektru. Namesto uporabe le enega od predstavljenih sledilnikov izboljšamo zmogljivost sledenja objektov v zaprtih prostorih z združitvijo obeh pristopov. To hkrati pomeni, da je implementiran sistem primeren za uporabo v povsem praktičnih aplikacijah, saj je načrtovan tako, da podpira sledenje do 16 gibajočih se objektov. V prvem delu naloge predstavimo obstoječe raziskovalno delo, ki je bilo opravljeno na tem področju, zatem opišemo podrobnosti implementacije in vrednotenja sistema ter na koncu predlagamo tudi množico direktiv za nadaljnje raziskovalno delo in možne izboljšave našega sistema.

• Language: Slovenian
• Keywords: računalniški vid, sledenje objektov, Kalmanov filter, filter z delci, Madžarska metoda, infrardeče oznake
• Work type: Master's thesis/paper
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2016
• PID: 20.500.12556/RUL-83680
• Publication date in RUL: 21.06.2016

Secondary language

• Language: English
• Title: Tracking system of known objects in unstable lightning conditions on stage using IR and visible spectrum video inputs

Abstract

In this research work we address the multiple object tracking problem under challenging illumination conditions. To solve the changing illumination problem, we implement two well known tracking approaches, which are operating in two separated domains. One tracker relies on active marker system, which operates in near infrared spectrum and the other one exploits existing state of the art pedestrian detector and color information by observing the scenery in visible spectrum. By integrating these two approaches, we intend to improve the performance of object tracking in closed areas, rather than using only one of the introduced trackers. This also means that the implemented system is usable in many practical applications, as it is designed to track up to 16 moving objects. In the first part of the thesis, we cover existing research work, which was done in this field, then we describe implementation and evaluation details of the system and finally we propose a set of directions for future research and possible improvements of our system.

• Keywords: computer vision, object tracking, Kalman filter, particle filter, Hungarian algorithm, infrared markers