Vizualno sledenje objektov s kvadrokopterjem

Muhovič, Jon Natanael (author), Kristan, Matej (mentor)

Zaradi vedno večje dostopnosti in kvalitete manjših letečih robotov za zasebno uporabo smo želeli nadgraditi njihovo funkcionalnost s sposobnostjo samodejnega sledenja želenemu objektu. To lahko naredimo z uporabo sledilnih algoritmov, ki delujejo na podlagi vizualne informacije. Taki algoritmi v vsaki sliki locirajo poljuben objekt in se prilagajajo spremembam velikosti in izgleda objekta. Formulirali smo sistem, ki uporabi informacije, pridobljene s takim algoritmom, za nadzor kvadrokopterja. Predlagani sistem tako ohranja razdaljo do sledenega objekta in hkrati skrbi, da je ta vedno blizu središča slike. Uporabili smo platformo s premično kamero, kar nam je omogočilo visoko odzivnost in sledenje objektu tudi, ko se ta nahaja niže od kvadrokopterja. Naš sistem nam poleg tega omogoča sledenje objektu, kadar se ta premika po neravnih površinah, kar s sistemi s fiksno kamero ni izvedljivo.

• Language: Slovenian
• Keywords: kvadrokopter, vizualno sledenje objektov, računalniški vid, teorija krmiljenja
• Work type: Master's thesis/paper
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2017
• PID: 20.500.12556/RUL-91373
• Publication date in RUL: 31.03.2017

Secondary language

• Language: English
• Title: Visual object tracking with a quadcopter

Abstract

Because of increasing accessibility and quality of small aerial robots for personal use we wanted to upgrade their functionality with the ability to autonomously follow objects. This can be achieved by using tracking algorithms that work with visual information. Such algorithms locate the object in every image and can adapt to size and appearance changes of the object. We have formulated a system that uses information from a tracking algorithm to control the quadcopter. The proposed system is thus able to maintain the distance to the object as well as keep the object near the center of the image. By using a platform with a movable camera we have achieved high responsiveness as well as the ability to follow an object that is located lower than the quadcopter itself. Our system also manages to follow targets that move on non-planar surfaces which is not possible using a fixed camera.

• Keywords: quadcopter, visual object tracking, computer vision, control theory