Matouš Vrba presents On Onboard LiDAR-based Flying Object Detection

Our new method for detecting flying objects using a 3D LiDAR sensor placed
onboard a UAV will be presented in a short seminar. This research is motivated
by the problem of autonomous physical aerial interception of UAVs intruding in
a
no-fly area by a specialized interceptor platform. For this purpose, a
low-latency, precise, and robust detection and relative localization of the
target using the interceptor's onboard sensors is crucial, which is tackled by
the proposed detection system. The main novelty of our approach is formulating
the detection problem within an occupancy mapping framework, which ensures a
low
rate of false detection of background objects as flying targets. A novel point
cloud-based multi-target tracking algorithm then provides further rejection of
sporadic false positives and unique identification, state estimation, and
trajectory prediction of the tracked objects. The resulting system fulfills the
strong requirements for implementing a practical autonomous aerial interception
system as we demonstrate in several real-world experiments. The system was also
deployed in other multi-robot scenarios, such as cooperative navigation and
exploration using UAVs.
Paper: https://doi.org/10.1109/TRO.2024.3502494
Videos, datasets: https://mrs.fel.cvut.cz/flying-object-detection
Code: https://github.com/ctu-mrs/vofod