US 12,072,439 B2
Synthetic generation of radar, LIDAR and ultrasound measurement data
Thomas Binzer, Ingersheim (DE); Anna Khoreva, Stuttgart (DE); and Juergen Hasch, Stuttgart (DE)
Assigned to ROBERT BOSCH GMBH, Stuttgart (DE)
Filed by Robert Bosch GmbH, Stuttgart (DE)
Filed on Oct. 22, 2020, as Appl. No. 17/077,947.
Claims priority of application No. 102019216927.9 (DE), filed on Nov. 4, 2019.
Prior Publication US 2021/0132189 A1, May 6, 2021
Int. Cl. G01S 7/40 (2006.01); G01S 7/41 (2006.01)
CPC G01S 7/4052 (2013.01) [G01S 7/412 (2013.01); G01S 7/417 (2013.01)] 13 Claims
OG exemplary drawing
 
1. A method for generating synthetic measurement data that is indistinguishable from actual measurement data captured by a first physical measurement modality, the first physical measurement modality being based on emitting an interrogating wave towards an object and recording a reflected wave coming from the object in a manner that allows for a determination of a time-of-flight between the emission of the interrogating beam and arrival of the reflected wave, the method comprising the following steps:
obtaining a first compressed representation of the synthetic measurement data in a first latent space, wherein the first latent space is associated with a first decoder that is trained to map each element of the first latent space to a record of synthetic measurement data that is indistinguishable from records of actual measurement data of the first physical measurement modality;
applying the first decoder to the first compressed representation, so as to obtain the synthetic measurement data; and
training, using the generated synthetic measurement data, at least one machine learning module that is to map actual measurement data captured from a vehicle to at least one classification and/or regression value, wherein the classification and/or regression value is used for operating the vehicle in road traffic in an at least partially automated manner;
wherein the obtaining of the first compressed representation includes:
applying a second trained encoder to a record of actual measurement data of a second physical measurement modality that is different from the first physical measurement modality, so as to obtain a second compressed representation of the actual measurement data of the second physical measurement modality in a second latent space, wherein the second latent space is associated with a second decoder that is trained to map each element of the second latent space to a record of synthetic measurement data that is indistinguishable from records of actual measurement data of the second physical measurement modality, and
applying a domain-transform, which is trained to map each element of the second latent space to an element of the first latent space, to the second compressed representation, so as to obtain the first compressed representation.