Flying Objects Detection from a Single Moving Camera

Supplementary material

Additional experimental results can be found here.

Try to find the drone in the image

The video below shows still frames and then a sequence of frames which makes it much easier to detect drones.

We can see that motion information plays a vital role here, as even humans have troubles in detecting drones in still images.

Comparison between:

  • Our detection algorithm based on the motion compensation algorithm ( green box )
  • Single-frame based method ( red box )
The single frame-based approach fails to detect the drone when it is in front of the trees, while our method can.

Comparing the performance of different approaches to motion compensation

  • Top row: Sequences of 4 consecutive frames before motion compensation.
  • Middle row: The same frames after motion compensation using our regression-based method.
  • Bottom row: The same frames after motion compensation using an optical-flow based motion method.

UAV examples
Aircraft examples

Aircrafts detection

Sample video filmed from the ground. The aircraft is correctly detected most of the time. The detector only fails when the plane is very close to the camera and moving too fast for the motion compensation algorithm.
Sample video recorded by the on-board camera of a radio-controlled plane filming another one flying at high speeds in front of complex backgrounds.