Tracking and precise landing by a quadrotor

• Main Author: Hoo, Shi Jan
• Other Authors: Ling Keck Voon
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2022
• Subjects: Engineering::Electrical and electronic engineering
• Online Access: https://hdl.handle.net/10356/158054
• Institution: Nanyang Technological University

For a drone to land precisely on a target in a flat terrain, it must accurately estimate its 4-dimensional (4D) position, which are the x-y-z coordinates and yaw orientation, with respect to its landing target. In this Final Year Project (FYP), we seek to investigate, adapt, and apply 4D pose estimation research for the purpose of drone landing on a ground robot in a flat terrain. A quadrotor and ground robot are simulated in a robot software environment in which the devised methodologies are implemented on the simulated drone to estimate its 4D pose with respect to the ground robot. The accuracy of the proposed implementation is measured by comparing the generated 4D pose estimates with the actual 4D pose values provided by the simulation. In an overview, the main methods used are deep learning for image points prediction, an efficient solution for the Perspective-N-Point (PNP) problem, and the Kalman filter for filtering / smoothening noisy estimates. These methods will be combined in tandem to form a pipeline for continuous pose estimation for the drone. In addition to this constructed pipeline, a section has been dedicated to exploratively address the problem of occlusion using Generative Adversarial Network, or in short GAN. Overall, the success of methods implemented within the simulation allows us to construct a proof-of-concept for 4D pose estimation on a real-life drone.