Enhanced Camera Target Tracking Using a Low-cost Thermal Camera

Abstract:

Visual cameras used in object tracking applications often struggle in poor visual
conditions sch as dark environments. This report investigates integration of a low-cost
thermal camera into the design of visual tracking systems to improve their performance
under poor visual conditions. This high level objective was then broken up into smaller
sub-objectives which aided with design processes such as component selection, mechanical
part designs and software development. All components were selected to provide the best
results within budget constraints with main components being the Adafruit AMG8833
Thermal camera, Raspberry Pi 5MP Mini camera and the Raspberry Pi 3B. The mechanical
parts were designed after the component selection process and the dimensions were
tailored to house the specific components. Two algorithms were developed, one that
uses only the visual camera to perform tracking and another one that uses both cameras.
The first algorithm was used as a benchmark to see how much the use of a thermal camera
can improve the performance of the system. The use of a thermal camera allowed the
device to track objects in darker environments where the visual camera performed poorly.
However the overall real-time tracking performance of both algorithms was poor due to
the computational limits of the Raspberry Pi 3B. Overall the device met all the basic
requirements set out at the start but can be improved a lot in future work

The full thesis document can be viewed here: OWSLEZ001_Thesis (1).pdf