Machine Learning for Countering UAV-Swarms
The proposed research objective is to detect UAVs from images taken by aircraft, ground vehicles, or persons by applying machine
learning models. The many different types of UAVs, such as micro-UAVs, large combat UAVs, GPS UAVs, High Altitude Long Endurance UAVs,
etc., can be differentiated by their visual looks. Efficient detection and identification of them can provide probable motives/missions
and how much priority should be given to them. A series of images of the UAVs can indicate velocity (direction and speed) but this approach
has multiple challenges. First, images are taken from different distances and resolutions and hence the distance of the UAVs from the camera
is estimated rather than fully known. Second, UAVs moving as, or within swarms/teams, have different formations when performing a group task
which complicates inference. Third, relative motion between the camera, the target, and the surrounding scene can cause a significant and
high-dynamic variation in illumination conditions, background characteristics, and target appearance. Therefore, three important detection
cases and techniques are posed as the subobjectives:
- To identify the UAV distance by detecting their size proportion in an image and their type with the actual size of a UAV;
- To detect the formations of the UAVs as a group; and
- Identification and classification of UAV by using radio frequency and electromagnetic emissions.
Researcher
|