Learning Visual Odometry for Unmanned Aerial Vehicles

Learning Visual Odometry for Unmanned Aerial Vehicles
Learning Visual Odometry for Unmanned Aerial Vehicles
Other Titles:
ICSIP 2017
Publication URL:
Publication Date:
04 August 2017
We describe a method to learn the flight dynamics of an unmanned aerial vehicle (UAV). This follows the recent trend to adopt a learning approach to Visual Odometry (VO). Our novelty is the inclusion of a module to compensate for the roll pitch angular motion of the UAV due to body vibration. This is a significant deviation from most existing works that are applied to land vehicles. We empirically verify our results on real flight data, showing that after compensating for angular vibration, the ego-motion of the UAV can be robustly estimated even by using simple regression tools. This enables the advantages of learning based VO to be within reach of the UAV community.
License type:
Funding Info:
(c) 2017 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other users, including reprinting/ republishing this material for advertising or promotional purposes, creating new collective works for resale or redistribution to servers or lists, or reuse of any copyrighted components of this work in other works.

Files uploaded: