Visual Object Detection and Tracking by the Crazyflie Quadcopter
Vizuální detekce a sledování objektu pomocí dronu Crazyflie
Authors
Supervisors
Reviewers
Editors
Other contributors
Journal Title
Journal ISSN
Volume Title
Publisher
České vysoké učení technické v Praze
Czech Technical University in Prague
Czech Technical University in Prague
Date
Abstract
V literární rešerši této práce byly prozkoumány současné i starší metody pro vizuální detekci objektů, sledování a odhad vzdálenosti. Byly také popsány hlavní složky ekosystému Crazyflie. V části implementace byly zavedeny vybrané metody. Pro detekci objektů byl použit model MobileNetV2-SSD s FPNlite, pro sledování objektů byl použit Kalmanův filtr a pro odhad vzdálenosti pomocí jediné kamery. Detekce objektů dosáhla průměrné přesnosti 87 % a přesnosti sledování 74 %. Hlavním výsledkem této práce je průzkum možného využití metod detekce a sledování objektů pomocí dronu Crazyflie 2.1 a jeho modulů Loco Positioning a AI-deck. V přílohách práce lze nalézt klíčové komponenty a finální implementaci létajícího streameru.
In this thesis's literature review section, contemporary and older methods for visual object detection, tracking, and distance estimation were investigated. The main components of the Crazyflie ecosystem were also described. In the implementation part, selected methods were implemented. For object detection, the MobileNetV2-SSD model with FPNlite was used, the Kalman Filter was employed for object tracking, and the monocular distance estimation technique was utilized for distance estimation. Object detection achieved an average accuracy of 87 % and a tracking accuracy of 74 %. This work's main result is exploring the potential use of object detection and tracking methods using the Crazyflie 2.1 drone and its Loco Positioning and AI-deck modules. In the appendices of the work, one can find key components and the final implementation of the flying streamer.
In this thesis's literature review section, contemporary and older methods for visual object detection, tracking, and distance estimation were investigated. The main components of the Crazyflie ecosystem were also described. In the implementation part, selected methods were implemented. For object detection, the MobileNetV2-SSD model with FPNlite was used, the Kalman Filter was employed for object tracking, and the monocular distance estimation technique was utilized for distance estimation. Object detection achieved an average accuracy of 87 % and a tracking accuracy of 74 %. This work's main result is exploring the potential use of object detection and tracking methods using the Crazyflie 2.1 drone and its Loco Positioning and AI-deck modules. In the appendices of the work, one can find key components and the final implementation of the flying streamer.