Augmentation and processing of radar data with machine learning

Patent metadata

Field	Value
Publication number	US-11899131-B2
Application number	US-202117391307-A
Country	US
Kind code	B2
Filing date	Aug 2, 2021
Priority date	Aug 28, 2020
Publication date	Feb 13, 2024
Grant date	Feb 13, 2024

Abstract

A method is disclosed for converting source radar data of a source configuration of a radar system target radar data of a target configuration. The method comprises: providing a source array of grid cells for source reflex locations; determining, for each respective grid cell in the source array, a probability or frequency that source reflex locations are located in the respective grid cell; forming a source tensor including the source array populated with the probability or frequency for each grid cell; transforming the source tensor into a target tensor including a target array of grid cells for the target reflex locations and indicating the probabilities or frequencies of the target reflex locations for each respective grid cell; and generating the target radar data by sampling the location coordinates of the target reflex locations.

First claim

What is claimed is:

1. A method for converting source radar data, obtained by observing a scene with a source configuration of a radar system, into target radar data that can be expected when the scene is observed with a target configuration of the radar system, the source radar data including location coordinates of source reflex locations from which reflected radar radiation was incident on the radar system, the target radar data including location coordinates of target reflex locations from which reflected radar radiation will be incident on the radar system, the method comprising:
providing a source array of grid cells for the location coordinates of the source reflex locations, the source array being one of two-dimensional and multi-dimensional;
determining, from the location coordinates of all of the source reflex locations, for each respective grid cell in the source array, at least one of a probability and a frequency that at least one source reflex location is located in and assigned to the respective grid cell;
forming a source tensor that includes the source array populated with the at least one of the probability and the frequency for each grid cell in the source array;
transforming the source tensor into a target tensor that includes at least one target array of grid cells for the location coordinates of the target reflex locations, the target array indicating, for each respective grid cell in the target array, at least one of a probability and a frequency that the target reflex locations obtained with the target configuration during observation of the scene will be located in the respective grid cell;
generating the target radar data by sampling, based on the at least one of the probability and the frequency for each grid cell in the target array, the location coordinates of the target reflex locations.

2. The method according to claim 1 , wherein:
the source configuration is identical to the target configuration; and
the source tensor is adopted identically as the target tensor.

3. The method according to claim 1 , wherein:
the source configuration differs from the target configuration; and
the source tensor is transformed into the target tensor using a trained machine learning model.

4. The method according to claim 3 , wherein the source configuration differs from the target configuration in that at least one of (i) the source configuration and the target configuration comprise different radar sensors, (ii) a radar sensor is spatially arranged differently in the source configuration compared to the target configuration, and (iii) the reflected radar radiation is influenced differently by materials arranged between a radar sensor and the observed scene, in the source configuration compared to the target configuration.

5. The method according to claim 3 , wherein the machine learning model has an encoder-decoder arrangement including (i) an encoder configured to map the source tensor onto a representation with reduced dimensionality and (ii) a decoder configured to map the representation onto the target tensor.

6. The method according to claim 3 , wherein the machine learning model has a generator of a generative adversarial network.

7. A method for processing source radar data, obtained by observing a scene with a source configuration of a radar system, the source radar data including location coordinates of source reflex locations from which reflected radar radiation was incident on the radar system, the method comprising:
providing a source array of grid cells for the location coordinates of the source reflex locations, the source array being one of two-dimensional and multi-dimensional;
determining, from the location coordinates of all of the source reflex locations, for each respective grid cell in the source array, at least one of a probability and a frequency that at least one source reflex location is located in and assigned to the respective grid cell;
forming a source tensor that includes the source array populated with the at least one of the probability and the frequency for each grid cell in the source array;
feeding the source tensor into a neural network.

8. The method according to claim 7 further comprising:
mapping, with the neural network, the source tensor onto at least one classes of a predefined classification.

9. The method according to claim 7 further comprising:
forming a control signal for a vehicle from an output supplied by the neural network; and
controlling the vehicle with the control signal.

10. The method according to claim 1 , wherein:
the source tensor includes assignments of at least one additional variable, which is derived from one of the source radar data and the target radar data, to the grid cells of the source array; and
the target tensor includes assignments of the at least one additional variable to the grid cells of the target array.

11. The method according to claim 10 further comprising:
forming, and attributing to a specific target reflex location of the target reflex locations in the target radar data, a new value of the at least one additional variable, based on at least one value of the at least one additional variable for at least one grid cell of the target array which have contributed to the sampling of the specific target reflex location.

12. The method according to claim 10 , wherein the at least one additional variable comprises at least one of:
a distance between one of (i) the radar system in the source configuration and the source reflex locations and (ii) the radar system in the target configuration and the target reflex locations;
an angle at which the reflected radar radiation one of (i) was incident on the radar system in the source configuration and (ii) will be incident on the radar system in the target configuration;
a speed of an object at which the reflected radar radiation was reflected;
an affiliation of an object at which the reflected radar radiation was reflected to at least one class of a predefined classification; and
a signal strength of the reflected radar radiation.

13. The method according to claim 1 , wherein one of a circle and a sphere with a predefined radius around one of the source reflex locations is assigned an occupancy probability of 1 and the occupancy probability is distributed over all grid cells of the source array over which one of an area of the circle and a volume of the sphere is distributed, according to one of area proportions and volume proportions respectively.

14. The method according to claim 1 , wherein the method is carried out by a computer program containing machine-readable instructions which are executed on at least one computer.

15. The method according to claim 14 , wherein the computer program is stored on a non-transitory machine-readable data carrier.

16. A computer having a non-transitory machine-readable data carrier that stores a computer program containing machine-readable instructions for converting source radar data, obtained by observing a scene with a source configuration of a radar system, into target radar data that can be expected when the scene is observed with a target configuration of the radar system, the source radar data including location coordinates of source reflex locations from which reflected radar radiation was incident on the radar system, the target radar data including location coordinates of target reflex locations from which reflected radar radiation will be incident on the radar system, the computer being configured to execute the machine-readable instructions to:
provide a source array of grid cells for the loc

Assignees

• Bosch Gmbh Robert

Inventors

• Junginger Andrej
• Oechsle Michael Johannes
• Strauss Thilo

Classifications

• G06N3/094

  Adversarial learning · CPC title

• G06N3/0455

  Auto-encoder networks; Encoder-decoder networks · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N3/0475

  Generative networks · CPC title

• G01S7/4082 Primary

  using externally generated reference signals, e.g. via remote reflector or transponder · CPC title