Small Drone Detection with Image Tiling in Counter Autonomous System Context

URL:

Abstract:

The increasing availability and versatility of drones in the last few years have made them an interesting tool to disrupt privacy, safety and security: they are small, fast and have sufficient payload to carry dangerous items. Because of their size and speed, they can be hard to detect and track. In this paper, we propose an efficient approach based on YOLOv3 [1] to detect drones in high resolution images by integrating an image tiling strategy. With this approach, we finished in second place in the 2020 Drone-vs-Bird Challenge (0.2% behind the winners). Finally, we also describe the ALEXIS system we have developed that can detect and classify using this approach, and track drones in real time from multiple cameras in different bands.

Attachment	Size
jdst_smalldronedetection_final_version-vol4-nom4.pdf	1.18 MB