Description:

This project aims to design and create a motion capture system that utilises multiple cameras to estimate the position and orientation of a rigid-body aerial robot in real time. This will be facilitated using computer vision and potentially also machine learning, such that the information from multiple cameras can be used in an appropriate algorithm, such as an extended Kalman filter, to provide accurate and robust state estimates of the rigid-body platform.

The intention is for the system to be low in cost, relative to off-the-shelf solutions, and also support ROS2, so that information from the camera system can be relayed to the aerial robot(s) of interest.

Objectives:

• Investigate the state of the art of optical tracking systems.
• Investigate camera models and computer vision.
• Select appropriate camera modules (and associated hardware) in terms of resolution, on-board processing, and cost.
• Design a suitable algorithm that is able to accurately estimate position and orientation of an aerial robot in real time.
• Simulate designed algorithm in Simulink.
• Implement algorithm on physical camera system and benchmark against off-the-shelf solutions.
• Write-up of dissertation.

Key skills/interests:

Mathematical modelling, machine vision, machine learning, state estimation

Expected outputs:

1x MSc (Eng), 1x Journal/Conference paper in a leading publication.

Supervisor:

Arnold Pretorius

Funding:

Student bursary funding may be available through a THRIP grant.