Frame-to-Frame ego-motion estimation for agile drones with Convolutional Neural Network

Keywords
Loading...
Thumbnail Image
Issue Date
2020-04-01
Language
en
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Openly available feature based Visual Odometery algorithms, have a high computational cost. This hinder the use of these algorithms in agile robotics platforms such as racing drones. A wide range of researchers have looked into, using data-driven learning methods such as Deep Learning, instead use of traditional feature based methods for developing a new class of visual odometery systems. These systems have mainly relied on datasets designed for low degree-of-freedom systems such as cars. Hence, even this new class of algorithms, su er from high computational cost and inability to handle agile robotics sytems. Therefore, this research will look into developing a learning based visual odometry system, speci cally designed for racing drones, which can handle high speed agile motion with low computational cost. Keywords| Drone, Neural Networks, Visual-odometry, Ego-motion
Description
Citation
Faculty
Faculteit der Sociale Wetenschappen