Method and device for generating synthetic training data for an artificial-intelligence machine for assisting with landing an aircraft

1. A computer-implemented method for generating synthetic training data for an artificial-intelligence machine, the method comprising at least steps of:

defining parameters for at least one approach scenario of an aircraft approaching a runway;

using the parameters of said at least one scenario in a flight simulator to generate simulated flight data, said flight simulator being configured to simulate said aircraft in the phase of approach toward said runway and to simulate an associated automatic pilot;

using the simulated flight data to generate a plurality of ground-truth images, said ground-truth images corresponding to various visibility conditions; and

generating, from each ground-truth image, a plurality of simulated sensor images, said simulated sensor images being able to be used as training data to train an artificial-intelligence machine for assisting with aircraft landing.

2. The method according to claim 1 , wherein the step of generating simulated sensor images consists in implementing, on a conditional generative adversarial neural network, an image-to-image translation algorithm, said image-to-image translation algorithm allowing a ground-truth image to be translated into a simulated sensor image.

3. The method according to claim 1 , wherein the step of generating simulated sensor images consists in generating simulated infrared images.

4. The method according to claim 1 , wherein the step of generating a plurality of ground-truth images comprises the steps of:

generating a first-level ground-truth image wherein only a restricted number of elements of interest feature in the image, each element of interest respectively being represented by one type of uniform information, said elements of interest comprising at least the runway represented by a first type of uniform information and the approach light system represented by a second type of uniform information; and

generating, from the first-level ground-truth image, various masked ground-truth images corresponding to different visibility conditions.

5. The method according to claim 1 , wherein each masked ground-truth image contains at least four classes or colours, each class respectively being associated with one label.

6. The method according to claim 5 , wherein at least four classes are the runway, the approach light system, visible data and invisible data.

7. The method according to claim 5 , wherein the step of generating simulated sensor images consists in translating a label in a ground-truth image into a texture for a simulated sensor image.

8. The method according to claim 1 , wherein the step of defining at least one approach scenario consists in defining parameters relative to at least one given runway, and to meteorological conditions and environmental conditions.

9. The method according to claim 1 , wherein the step of generating simulated flight data consists in simulating various approach paths of the simulated aircraft and of the automatic pilot with the scenario parameters, and in generating simulated flight data for the various paths comprising the position (latitude, longitude, altitude) and the attitude (yaw, pitch, roll) of the simulated aircraft.

10. The method according to claim 1 , in addition comprising a step consisting in storing the plurality of simulated sensor images in a database of synthetic training data.

11. A device for generating synthetic training data for an artificial-intelligence machine, the device comprising means for implementing the steps of the method according to claim 1 .

12. A computer program containing code instructions for executing the steps of the method for generating synthetic training data for an artificial-intelligence machine, according to claim 1 , when said program is executed by a processor.