Methods and systems for detection of objects in a vicinity of a vehicle

Patent metadata

Field	Value
Publication number	US-12352849-B2
Application number	US-202117384493-A
Country	US
Kind code	B2
Filing date	Jul 23, 2021
Priority date	Jul 24, 2020
Publication date	Jul 8, 2025
Grant date	Jul 8, 2025

Abstract

A computer implemented method for detection of objects in a vicinity of a vehicle comprises the following steps carried out by computer hardware components: acquiring radar data from a radar sensor; determining a plurality of features based on the radar data; providing the plurality of features to a single detection head; and determining a plurality of properties of an object based on an output of the single detection head.

First claim

What is claimed is:

1. A method, comprising:
detecting, using computer hardware components of a vehicle, objects in a vicinity of a vehicle by:
acquiring radar data from a radar sensor;
determining a plurality of features based on the radar data;
providing the plurality of features to a single detection head that comprises a plurality of sequentially arranged layers, the single detection head being free from layers arranged in parallel;
determining a plurality of properties of each object based on an output of the single detection head;
carrying out, with an ego-motion compensation module, a nearest neighbor interpolation to determine a new position of each object in a current time step and avoiding drift due to an accumulation of positional errors in positions of the objects over time by recording a residual part indicating a positional error of a movement of each object from the current time step;
determining the new position of each object based on a transformation grid from the current time step and a previous residual part for each object from a previous time step;
determining a subsequent position of each object in a subsequent time step based on the recorded residual part for each object; and
detecting the objects based on a regression subnet comprising the ego-motion compensation module; and

controlling, by an autonomous driving system of the vehicle, autonomous driving of the vehicle based on the plurality of properties of each object.

2. The method of claim 1 ,
wherein each of the plurality of features is connected to an input of the single detection head.

3. The method of claim 1 ,
wherein the plurality of features are determined using an artificial neural network.

4. The method of claim 1 ,
wherein the single detection head is trained for the plurality of properties simultaneously.

5. The method of claim 1 ,
wherein the plurality of properties comprises at least two of a class of the object, a size of the object, or a yaw angle of the object.

6. The method of claim 1 , wherein detecting the objects in the vicinity of the vehicle further comprises:
determining a radar data cube based on the radar data;
providing the radar data cube to a plurality of layers of a neural network;
resampling the output of the plurality of layers into a vehicle coordinate system; and
determining the plurality of features based on the resampled output.

7. The method of claim 1 , wherein detecting the objects in the vicinity of the vehicle further comprises:
fusing data from a plurality of radar sensors including the radar sensor; and
determining the plurality of features further based on the fused data.

8. The method of claim 1 , wherein detecting the objects in the vicinity of the vehicle further comprises:
acquiring camera data from a camera;
wherein the plurality of features are determined further based on the camera data.

9. The method of claim 1 , wherein detecting the objects in the vicinity of the vehicle further comprises:
acquiring lidar data from a lidar sensor; and
determining the plurality of features further based on the lidar data.

10. The method of claim 1 , wherein detecting the objects in the vicinity of the vehicle further comprises determining an angle of arrival based on the radar data.

11. The method of claim 10 , wherein the angle of arrival is determined using an artificial neural network with a plurality of layers.

12. The method of claim 11 , wherein the artificial neural network further comprises a dropout layer.

13. The method of claim 1 , wherein the regression subnet further comprises at least one of a u-shaped network, or a LSTM.

14. The method of claim 1 , wherein the ego-motion compensation module is configured to carry out ego-motion compensation of an output of a recurrent network of the previous time step, and input the result of the ego-motion compensation into a recurrent network of the current time step.

15. A system comprising a plurality of computer hardware components configured to:
detect objects in a vicinity of a vehicle by:
acquiring radar data from a radar sensor;
determining a plurality of features based on the radar data;
providing the plurality of features to a single detection head that comprises a plurality of sequentially arranged layers, the single detection head being free from layers arranged in parallel;
determining a plurality of properties of each object based on an output of the single detection head;
carrying out, with an ego-motion compensation module, a nearest neighbor interpolation to determine a new position of each object in a current time step and avoiding drift due to an accumulation of positional errors in positions of the objects over time by recording a residual part indicating a positional error of a movement of each object from the current time step;
determining the new position of each object based on a transformation grid from the current time step and a previous residual part for each object from a previous time step;
determining a subsequent position of each object in a subsequent time step based on the recorded residual part for each object; and
detecting the objects based on a regression subnet comprising the ego-motion compensation module; and

control autonomous driving of the vehicle based on the plurality of properties of each object.

16. A vehicle comprising the system of claim 15 .

17. A non-transitory computer readable medium comprising instructions that, when executed, configure a plurality of computer hardware components to:
detect objects in a vicinity of a vehicle by:
acquiring radar data from a radar sensor;
determining a plurality of features based on the radar data;
providing the plurality of features to a single detection head that comprises a plurality of sequentially arranged layers, the single detection head being free from layers arranged in parallel;
determining a plurality of properties of each object based on an output of the single detection head;
carrying out, with an ego-motion compensation module, a nearest neighbor interpolation to determine a new position of each object in a current time step and avoiding drift due to an accumulation of positional errors in positions of the objects over time by recording a residual part indicating a positional error of a movement of each object from the current time step;
determining the new position of each object based on a transformation grid from the current time step and a previous residual part for each object from a previous time step;
determining a subsequent position of each object in a subsequent time step based on the recorded residual part for each object; and
detecting the objects based on a regression subnet comprising the ego-motion compensation module; and

control autonomous driving of the vehicle based on the plurality of properties of each object.

18. The system of claim 15 , wherein each of the features is connected to an input of the single detection head.

19. The system of claim 15 , wherein the single detection head is trained for the plurality of properties simultaneously.

Assignees

• Aptiv Technologies AG

Inventors

• Meuter Mirko
• Kurian Jittu
• Su Yu
• Siegemund Jan
• Niu Zhiheng
• Lessmann Stephanie
• Khalili Dehkordi Saeid
• Kästner Florian
• Kossaczky Igor
• Labusch Sven
• Grumpe Arne
• Schoeler Markus
• Luszek Moritz
• Zhu Weimeng
• Becker Adrian
• Cennamo Alessandro
• Kollek Kevin
• Braun Marco
• Spata Dominic
• Roesler Simon

Classifications

• G06N3/0495

  Quantised networks; Sparse networks; Compressed networks · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N3/0442

  characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU] · CPC title

• G06N3/096

  Transfer learning · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title