We propose an approach to detect flying objects such as UAVs and aircrafts when
they occupy a small portion of the field of view,
possibly moving against complex backgrounds, and are filmed by a camera that
itself moves.
Solving such a difficult problem requires combining both appearance and motion
cues. To this end we propose a regression-based approach
to motion stabilization of local image patches that allows us to achieve
effective classification on spatio-temporal image cubes and
outperform state-of-the-art techniques.

As the problem is relatively new and there is very limited amount of real aerial
footage available, we further propose a way to
synthetically generate image patches. The latter are designed to be similar to
the real samples from the point of view of the detector.
This leaves us with a relatively small real dataset and large amount of
synthetic samples, which increases the risk of over-fitting to
synthetic data, especially when using Deep Convolutional Networks. In this
context we propose an effective way to combine real and
generated synthetic samples to avoid such a problem.

Artem is working at EPFL pursuing his Ph.D. in Computer Vision in the field of
detection of small fast moving objects from a single
moving camera under the supervision of Prof. Pascal Fua and Prof. Vincent
Lepetit. Artem is currently spending two months as an intern at
Honeywell in Prague.