Alexander Böckenholt
FPGA-based Object Detection for Control of Autonomous UAVs
Abstract
There are many approaches to drone control based on object detection. However, these
approaches often prioritize performance and generality over more resource efficient
alternatives. In most cases, the situations allows for some preparation of the environment
which significantly lowers the resource requirements of systems to accomplish
most task. In this work an implementation was designed, which forms a low-cost base
system that covers the control of a drone based on object detection. To achieve this,
it uses markers, which are designed detected, as a reference point for the navigation.
The object detection which determines the position is implemented on an FPGA and
optimized for minimal resource cost. In the tests the control program displayed the
ability to enact fine movements on the drone. The implemented object detection was
able to be implemented with only 7566 FFs and 9083 LUTs and still achieve a frame
rate of 26.78 frames per second. The system and the internal implementation are both
modular to allow the fast adaptation to application specific circumstances. This works
implementation of the complete system is not functional, because of a problem in the
communication between PS and PL. Despite that it forms the blueprint for a solid base
architecture for UAV applications that already manages its control.