Development of a drone positioning system for tiled walls
Project Description

There are several startups worldwide focusing on parcel delivery service. While the Global Positioning System allows a drone to fly to a specified outdoor location, it is not accurate enough for a drone to position itself to a specific window located at a specified floor. A new vision positioning solution using the patterns of tiled walls in building exteriors has been devised by a HKUST team to address this challenge. A proof-of-concept demo has been developed to show its feasibility. However, the existing Matlab-/Python-based prototype is slow and not so stable. The goal of this project is to understand the vision positioning approach, adapt the algorithm for CUDA C/C++ implementation in an embedded GPU board (Nvidia’s Jetson Nano), and develop a reliable system demo using a stable drone (e.g. DJI Mavic 2 Zoom).
Prerequisite: Good software skill and know Matlab/Python/C/C++.

Supervisor
MOW Wai Ho
Quota
5
Course type
UROP1000
UROP1100
Applicant's Roles

1) Adapt the Matlab-/Python-based program to CUDA C/C++ implementation in an embedded GPU board;
2) Develop a reliable system demo using a stable drone.

Applicant's Learning Objectives

1) Understand the advantages and limitations of the vision positioning approach;
2) Learn GPU programming using CUDA C/C++ and how it can be used to realize a real-time system;
3) Learn how to develop a reliable drone positioning solution using an embedded system approach.

Complexity of the project
Moderate