Skip to main content

Simulation · Photorealistic

Exercises with real imagery

Dome Lite gives you contacts. A game engine gives you pixels, which is what a camera detector has to be tested against.

Why it matters​

Every sensor except a camera can be simulated convincingly from geometry. A radar return is a range and a bearing with noise on it. An RF bearing is an angle with noise on it. Those models are honest, and Dome Lite runs them.

A camera is different. What a camera contributes is identity, and identity comes from pixels. You cannot test a detector against a simulated bounding box, because the detector's whole job is producing the bounding box in the first place.

So a photorealistic environment is what lets you answer questions Dome Lite cannot:

  • At what range does the detector actually recognise a quadcopter?
  • What does it do at dusk, in rain, against a treeline?
  • How much does a lens smear at closing speed cost you?
  • Does the identity channel firm up early enough to change a decision?

What it changes and what it does not​

The aircraft in a game-engine environment are real autopilots running against a simulated airframe. They speak the same protocol a real drone speaks, so nothing downstream of the connection is different. The picture, fusion, the planner and the gates all behave exactly as they do live.

That is the reason this is worth doing properly rather than as a graphics feature. The pipeline does not know it is a simulation.

Where to go next​