Valeo use a NN to determine areas of interest:
EP4456014A1 METHOD TO OPERATE AN ADAPTIVE HEADLIGHT DEVICE OF A VEHICLE 20230428
[0057] The
artificial neural network 24 is preferably a convolutional neural network (CNN), in particular, a multi-branch CNN. Further sub-steps of step S4 are possible which are here indicated as steps S5 to S9. The artificial neural network 24 may comprise three separate input branches 16, 17, 18 for feature extraction. The provide outside information 10 is provided to a first input branch 16 of the three input branches 16, 17, 18 in step S5. The provided weather information 14 is provided to a second input branch 17 in step S6, and the lighting information 15 is provided to a third input branch 18 of the three input branches 16, 17, 18 in step S7. The second input branch 17 and/or the third input branch 18 are preferably fed with data determined by the further artificial neural networks 23. To determine the weather information 14 and/or the lighting information 15, the respective further artificial neural network 23 may be applied on the provided outside information 10 and/or on the further sensor information. The further sensor information may be provided by the rain sensor 5, the temperature sensor 6 and/or by the external device 9. There may be only one single artificial neural network 24 that determines the weather information 14, the lighting information 15 and the control command 20. Otherwise, the artificial neural network 24 and the two further artificial neural networks 23 differ from each other.
[0060] Applying the artificial neural network 24 may comprise
determining at least one area of interest in the environment 3 of the vehicle 1. The area of interest may comprise at least partially an opposing vehicle 1, a traffic sign, an obstacle and/or a road infrastructure element. It is hence possible to define which area in the environment 3 should be more or less illuminated than at a current point in time.
Given the filing date is 2023, it is possible that they use Akida/TENNs software to determine the areas of interest.
for that , .......... though..... we may see or hear something intriguing yet.
