Partially-learned model for speed estimates in radar tracking

Patent metadata

Field	Value
Publication number	US-12092734-B2
Application number	US-202117644464-A
Country	US
Kind code	B2
Filing date	Dec 15, 2021
Priority date	Mar 25, 2021
Publication date	Sep 17, 2024
Grant date	Sep 17, 2024

Abstract

This document describes techniques and systems for a partially-learned model for speed estimates in radar tracking. A radar system is described that determines radial-velocity maps of potential detections in an environment of a vehicle. The model uses a data cube to determine predicted boxes for the potential detections. Using the predicted boxes, the radar system determines Doppler measurements associated with the potential detections that correspond to the predicted boxes. The Doppler measurements are used to determine speed estimates for the predicted boxes based on the corresponding potential detections. These speed estimates may be more accurate than a speed estimate derived from the data cube and the model. Driving decisions supported by the speed estimates may result in safer and more comfortable vehicle behavior.

First claim

What is claimed is:

1. A method, comprising:
generating, from radar signals, a data cube that represents an environment of a vehicle;
determining, based on the radar signals or the data cube, radial-velocity maps of potential detections in the environment of the vehicle, the radial-velocity maps including radial velocity data and Doppler data for the potential detections;
determining, using a model applied to the data cube, predicted boxes corresponding to respective groups of the potential detections, the predicted boxes including at least modeled speed estimates associated with each of the predicted boxes;
determining Doppler measurements associated with groups of the potential detections associated with each of the predicted boxes, the Doppler measurements including, for each of the box predictions, a list of measured radial velocities produced independent from an output from the model such that the list of measured radial velocities are associated with the groups of the potential detections determined for each of the box predictions;
determining, using the Doppler measurements associated with the groups of the potential detections determined for each of the predicted boxes, improved speed estimates for each of the predicted boxes by fusing ones of the corresponding measured radial velocities that align with areas of the box predictions generated using the model to replace the modeled speed estimates from the data cube with the improved speed estimates determined from the measured radial velocities; and
providing the improved speed estimates for the predicted boxes as an input to an autonomous-driving system or an assisted-driving system that operates the vehicle on a roadway.

2. The method of claim 1 , wherein determining of the radial-velocity maps of the potential detections in the environment of the vehicle is performed by the model.

3. The method of claim 1 , wherein the data cube includes at least three dimensions, the at least three dimensions including at least range data, angle data, and radial velocity data for the potential detections.

4. The method of claim 1 , the method further comprising:
performing a polar to cartesian transformation on the radial-velocity maps of the potential detections resulting in transformed radial-velocity maps; and
generating a fusion of coordinates for the transformed radial-velocity maps, the fusion of coordinates maintaining a radial-velocity value with a largest magnitude for a given angle for each of the potential detections.

5. The method of claim 1 , wherein the model comprises a neural network, a convolutional neural network, a recurrent neural network, a modular neural network, or a long short-term memory network.

6. The method of claim 1 , further comprising:
compensating the measured radial velocities based on a speed of the vehicle to determine a compensated range rate for the predicted boxes.

7. The method of claim 6 , wherein the improved speed estimates are determined using a least-squares algorithm to determine a cartesian speed that best matches the Doppler measurements.

8. The method of claim 7 , wherein the improved speed estimates are also determined using an orientation of the predicted boxes as predicted by the model and the compensated range rate.

9. The method of claim 1 , wherein the radar signals are generated by a radar system that is configured to be installed on an automobile.

10. A radar system comprising one or more processors configured to:
generate, from radar signals, a data cube that represents an environment of a vehicle;
determine, based on the radar signals or the data cube, radial-velocity maps of potential detections in the environment of the vehicle, the radial-velocity maps including radial velocity data and Doppler data for the potential detections;
determine, using a model applied to the data cube, predicted boxes corresponding to respective groups of the potential detections, the predicted boxes including at least modeled speed estimates associated with each of the predicted boxes;
determine Doppler measurements associated with groups of the potential detections associated with each of the predicted boxes, the Doppler measurements including, for each of the box predictions, a list of measured radial velocities produced independent from an output from the model such that the list of measured radial velocities are associated with the groups of the potential detections determined for each of the box predictions;
determine, using the Doppler measurements associated with the groups of the potential detections determined for each of the predicted boxes, improved speed estimates for each of the predicted boxes by fusing ones of the corresponding measured radial velocities that align with areas of the box predictions generated using the model to replace the modeled speed estimates from the data cube with the improved speed estimates determined from the measured radial velocities; and
provide the improved speed estimates for the predicted boxes as an input to an autonomous-driving system or an assisted-driving system that operates the vehicle on a roadway.

11. The radar system of claim 10 , wherein a determination of the radial-velocity maps of the potential detections in the environment of the vehicle is performed by the model.

12. The radar system of claim 10 , wherein the data cube includes at least three dimensions, the at least three dimensions including at least range data, angle data, and radial velocity data for the potential detections.

13. The radar system of claim 10 , the one or more processors further configured to:
perform a polar to cartesian transformation on the radial-velocity maps of the potential detections resulting in transformed radial-velocity maps; and
generate a fusion of coordinates for the transformed radial-velocity maps, the fusion of coordinates maintaining a radial-velocity value with a largest magnitude for a given angle for each of the potential detections.

14. The radar system of claim 10 , wherein the model comprises a neural network, a convolutional neural network, a recurrent neural network, a modular neural network, or a long short-term memory network.

15. A non-transitory computer-readable medium comprising instructions that, when executed, cause one or more processors of a radar system to:
generate, from radar signals, a data cube that represents an environment of a vehicle;
determine, based on the radar signals or the data cube, radial-velocity maps of potential detections in the environment of the vehicle, the radial-velocity maps including radial velocity data and Doppler data for the potential detections;
determine, using a model applied to the data cube, predicted boxes corresponding to respective groups of the potential detections, the predicted boxes including at least modeled speed estimates associated with each of the predicted boxes;
determine Doppler measurements associated with groups of the potential detections associated with each of the predicted boxes, the Doppler measurements including, for each of the box predictions, a list of measured radial velocities produced independent from an output from the model such that the list of measured radial velocities are associated with the groups of the potential detections determined for each of the box predictions;
determine, using the Doppler measurements associated with the groups of the potential detections determined for each of the predicted boxes, improved speed estimates for each of the predicted boxes by fusing ones of the corresponding measured radial velocities that align with areas of the box predictions generated us

Assignees

• Aptiv Technologies AG

Inventors

• Roesler Simon
• Becker Adrian
• Schiffmann Jan K

Classifications

• G06N3/02

  Neural networks · CPC title

• G01S13/89

  for mapping or imaging · CPC title

• G01S13/58

  Velocity or trajectory determination systems; Sense-of-movement determination systems · CPC title

• G01S7/415

  Identification of targets based on measurements of movement associated with the target · CPC title

• G01S13/726

  Multiple target tracking · CPC title