Methods for classifying objects in automotive-grade radar signals
A method includes an operation to collect radar signals reflected from objects in a field of view. A three-dimensional range-angle-velocity cube is formed from the radar signals. The three-dimensional range-angle-velocity cube includes individual bins with radar intensity values characterizing angle and range for a specific velocity. Point-pillar sub-cubes are selected from the three-dimensional range-angle-velocity cube. Each point-pillar sub-cube includes a predefined range surrounding a high energy peak in the range-angle dimensions and an entire range in a velocity vector. The point-pillar sub-cubes are processed to compress, decompress, detect, classify or track objects in the field of view.
1. A method, comprising:
collecting radar signals reflected from objects in a field of view;
forming a three-dimensional range-angle-velocity cube from the radar signals, the three-dimensional range-angle-velocity cube including individual bins with radar intensity values characterizing angle and range for a specific velocity;
selecting point-pillar sub-cubes from the three-dimensional range-angle-velocity cube, each point-pillar sub-cube including a region surrounding a high energy peak in range-angle dimensions and an entire range in a velocity vector; and
processing the point-pillar sub-cubes to compress, decompress, detect, classify or track objects in the field of view including:
compressing velocity data in the point-pillar sub-cubes using one or more of a neural network or a pointwise operator before applying a classifier; and
reconstructing uncompressed velocity data in the point-pillar sub-cubes.
2. The method of claim 1 , wherein the region is defined by having values within a specified percentage of the high energy peak.
3. The method of claim 1 , wherein the region is defined by being within a threshold distance of the high energy peak.
4. The method of claim 1 , wherein the compressing the velocity data uses the pointwise operator.
5. The method of claim 1 , wherein the compressing the velocity data uses the neural network.
6. The method of claim 1 , wherein the reconstructing uncompressed velocity data occurs after classification in a processing pipeline.
7. The method of claim 1 wherein the neural network is used to detect, classify or track objects in the field of view.
8. The method of claim 1 , wherein a compressed representation of the point-pillar sub-cubes are transmitted to another processing facility where classification is performed.
9. A radar system, comprising:
a transmitter configured to transmit radar signals;
a receiver configured to receive radar signals reflected from objects in a field of view;
a signal processor configured to:
form a three-dimensional range-angle-velocity cube from the received radar signals, the three-dimensional range-angle-velocity cube including individual bins with radar intensity values characterizing angle and range for a specific velocity;
select point-pillar sub-cubes from the three-dimensional range-angle-velocity cube, each point-pillar sub-cube including a region surrounding a high energy peak in range-angle dimensions and an entire range in a velocity vector; and
process the point-pillar sub-cubes to compress, decompress, detect, classify or track objects in the field of view including:
compressing velocity data in the point-pillar sub-cubes using one or more of a neural network or a pointwise operator before applying a classifier; and
reconstructing uncompressed velocity data in the point-pillar sub-cubes.
10. The radar system of claim 9 , wherein the region is defined by having values within a specified percentage of the high energy peak.
11. The radar system of claim 9 , wherein the region is defined by being within a threshold distance of the high energy peak.
12. The radar system of claim 9 , wherein the compressing the velocity data uses the pointwise operator.
13. The radar system of claim 9 , wherein the compressing the velocity data uses the neural network.
14. The radar system of claim 9 , wherein the signal processor utilizes the neural network to detect, classify or track objects in the field of view.
15. The radar system of claim 9 , wherein the transmitter is configured to transmit frequency modulated continuous wave (FMCW) signals.
16. The radar system of claim 9 , further comprising a duplexer configured to isolate the transmitted and received radar signals.
17. The radar system of claim 9 , wherein the signal processor is configured to trigger transmission of a compressed representation of the point-pillar sub-cubes to another processing facility where classification is performed.