Methods and systems for object detection

Patent metadata

Field	Value
Publication number	US-11604272-B2
Application number	US-202016904835-A
Country	US
Kind code	B2
Filing date	Jun 18, 2020
Priority date	Jul 18, 2019
Publication date	Mar 14, 2023
Grant date	Mar 14, 2023

Abstract

A computer implemented method for object detection includes: determining a grid, the grid comprising a plurality of grid cells; determining, for a plurality of time steps, for each grid cell, a plurality of respective radar detection data, each radar detection data indicating a plurality of radar properties; determining, for each time step, a respective radar map indicating a pre-determined radar map property in each grid cell; converting the respective radar detection data of the plurality of grid cells for the plurality of time steps to a point representation of pre-determined first dimensions; converting the radar maps for the plurality of time steps to a map representation of pre-determined second dimensions, wherein the pre-determined first dimensions and the pre-determined second dimensions are at least partially identical; concatenating the point representation and the map representation to obtain concatenated data; and carrying out object detection based on the concatenated data.

First claim

We claim:

1. A computer implemented method for object detection, the method comprising:
determining, by at least one processor, for a plurality of time steps, for each of a plurality of grid cells of a grid, a plurality of respective radar detection data, each radar detection data indicating a plurality of radar properties;
determining, by the at least one processor, for each time step, a respective radar map indicating a pre-determined radar map property in each grid cell;
converting, by the at least one processor, the respective radar detection data of the plurality of grid cells for the plurality of time steps to a point representation of pre-determined first dimensions;
converting, by the at least one processor, the radar maps to a map representation of pre-determined second dimensions, wherein the pre-determined first dimensions and the pre-determined second dimensions are at least partially identical;
concatenating, by a neural network, the point representation and the map representation to obtain concatenated data; and
carrying out object detection based on the concatenated data.

2. The computer implemented method of claim 1 , comprising:
receiving radar sensor data for each time step from a radar sensor provided on a vehicle; and
preprocessing the radar sensor data to remove an effect of a change in location of the vehicle;

wherein the radar detection data and the radar maps are determined based on the preprocessed radar sensor data.

3. The computer implemented method of claim 2 , wherein, for each time step, the radar sensor data is preprocessed to remove the effect of the change in location of the vehicle based on the radar sensor data of the time step and radar sensor data of a pre-determined number of previous time steps preceding the time step.

4. The computer implemented method of claim 1 , wherein, for each time step, the respective radar map comprises a motion map indicating a probability of existence of a radar point detection at a next time step.

5. The computer implemented method of claim 1 , wherein, for each time step, the respective radar map comprises a positional uncertainty map indicating a positional uncertainty at the respective time step.

6. The computer implemented method of claim 1 , wherein, for each time step, the respective radar map comprises an occupancy map indicating static object contours.

7. The computer implemented method of claim 1 , wherein converting the respective radar detection data of the plurality of grid cells for the plurality of time steps to the point representation of pre-determined first dimensions comprises reducing a dimension of points in each cell.

8. The computer implemented method of claim 7 , wherein reducing the dimension of points comprises max pooling.

9. The computer implemented method of claim 1 ,
wherein the grid comprises a pre-determined height and a pre-determined width,
the point representation comprises a four-dimensional point tensor,
a first dimension of the point tensor corresponds to the number of the plurality of time steps,
a second dimension of the point tensor corresponds to the height of the grid,
a third dimension of the point tensor corresponds to the width of the grid,
the map representation comprises a four-dimensional map tensor,
a first dimension of the map tensor corresponds to the number of the plurality of time steps,
a second dimension of the map tensor corresponds to the height of the grid, and
a third dimension of the map tensor corresponds to the width of the grid.

10. The computer implemented method of claim 9 , wherein
the concatenated data comprises a four-dimensional concatenated tensor, a first dimension of the concatenated tensor corresponds to the number of the plurality of time steps,
a second dimension of the concatenated tensor corresponds to the height of the grid,
a third dimension of the concatenated tensor corresponds to the width of the grid, and
a fourth dimension of the concatenated tensor corresponds to the sum of a fourth dimension of the point tensor and a fourth dimension of the map tensor.

11. The computer implemented method of claim 9 , comprising reducing the
dimension of the concatenated data to get time fused data, and

wherein
the time fused data comprises a three-dimensional time fused tensor,
a first dimension of the time fused tensor corresponds to the height of the grid,
a second dimension of the time fused tensor corresponds to the width of the grid,
a third dimension of the time fused tensor corresponds to the sum of a fourth dimension of the point tensor and a fourth dimension of the map tensor, and
object detection is carried out based on the time fused data.

12. The computer implemented method of claim 11 , wherein reducing the dimension of the concatenated data comprises using a 3D convolution.

13. The computer implemented method of claim 11 , wherein reducing the dimension of the concatenated data comprises using a recurrent network.

14. A computer system comprising a plurality of computer hardware components configured to carry out the computer implemented method of claim 1 .

15. A non-transitory computer readable medium comprising instructions for carrying out the computer implemented method of claim 1 .

Assignees

• Aptiv Tech Ltd

Inventors

• Su Yu
• Zhu Weimeng
• Kästner Florian
• Becker Adrian

Classifications

• G01S13/931 Primary

  of land vehicles · CPC title

• G01S7/415

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

• G06N3/045

  Combinations of networks · CPC title

• G01S2013/932

  using own vehicle data, e.g. ground speed, steering wheel direction · CPC title

• G01S7/417

  involving the use of neural networks · CPC title