Visual drone detection pilot

Project at a glance

Client

Deutsche Telekom AG

Period

April to July 2017

Team

3 people: 1 developer, 1 AI specialist, 1 project manager

Technologies

• Convolutional neural networks (CNN)
• Classical image recognition

Senva's role

Pilot for visual detection

Starting point

With its drone shield, Deutsche Telekom set up a programme to detect drones. The pilot was meant to establish how well drones can be detected from camera images alone.

Senva's role

Senva developed a pilot for visual detection: software that finds drones in camera images. Senva did not develop the drone shield programme itself. Our contribution was this one component.

Approach

We built a hybrid model, in which a convolutional neural network works together with classical image recognition methods.

Looking back, we would still recommend this approach when training data is scarce. In that situation, a learned model on its own is rarely the most reliable choice.

Result

The pilot was completed successfully and helped improve drone detection within the programme.