Detecting Object Movement from a Moving UAV Video Stream
The aim of the work is to develop a system which can reliably detect movements in an observed area with the help of the integrated camera of an omnidirectional flight system.
Although this domain is already well researched, there is a lack of publicly available data sets that are needed to train reliable neural networks.
Therefore, the question is whether it is possible to use simulated data that has been generated purely virtually to train a system that can then deliver comparable results in the real world.
For this purpose, data sets were first generated with various freely available game engines. Then a neural network was trained on this self-generated data.
Finally, the trained model was validated on existing data and the results were compared with those of previous work.
The trained system is finally able to compete on pair with current technology and at the same time meet real-time requirements. It is possible to detect movement in a monitored area with the system, even when the aircraft is moving.