	US 12,190,242 B2
	Methods for detecting phantom projection attacks against computer vision algorithms
Ben Nassi, Ramat-Gan (IL); Yuval Elovici, Arugot (IL); Yisroel Avraham Mirsky, Beer Sheva (IL); Dudi Nassi, Holon (IL); and Raz Ben Nethanel, Petah Tikva (IL)
Assigned to B. G. NEGEV TECHNOLOGIES AND APPLICATIONS LTD., Beer Sheva (IL)
Appl. No. 17/779,774
Filed by B. G. NEGEV TECHNOLOGIES AND APPLICATIONS LTD., Beer Sheva (IL)
PCT Filed Nov. 25, 2020, PCT No. PCT/IL2020/051211 § 371(c)(1), (2) Date May 25, 2022, PCT Pub. No. WO2021/105985, PCT Pub. Date Jun. 3, 2021.
Claims priority of provisional application 62/940,280, filed on Nov. 26, 2019.
Prior Publication US 2023/0015771 A1, Jan. 19, 2023
Int. Cl. G06N 3/08 (2023.01); G06V 10/25 (2022.01); G06V 10/26 (2022.01); G06V 10/82 (2022.01); G06V 20/00 (2022.01); G06V 20/56 (2022.01); G06V 20/58 (2022.01); B60W 60/00 (2020.01)

CPC G06N 3/08 (2013.01) [G06V 10/25 (2022.01); G06V 10/267 (2022.01); G06V 10/82 (2022.01); G06V 20/56 (2022.01); G06V 20/58 (2022.01); G06V 20/95 (2022.01); B60W 60/001 (2020.02); B60W 2420/403 (2013.01); B60W 2554/00 (2020.02); G06V 2201/08 (2022.01)]

20 Claims

1. A method for protecting a vehicle employing an advanced driver-assistance system (ADAS) from a phantom projection attack, the method comprising:

receiving an image from an image sensor, wherein the received image includes surroundings of the vehicle;

processing the received image to identify a traffic object in the image;

creating multiple processed images from the received image, wherein each processed image highlights a different distinct aspect of the image;

applying the multiple processed images to multiple respective neural network (NN) aspect models trained to correlate the distinct aspect of each processed image with a phantom projection attack;

applying latent representations of the multiple processed images from each of the NN aspect models to a NN combiner model trained to determine whether the latent representations indicate a phantom projection attack; and

responsively to a determination by the NN combiner model that there is a phantom projection attack, issuing a phantom projection indicator to a decision module of the ADAS.