Radar data processing using neural network classifier and confidence metrics
A radar data processing device includes at least one analog-to-digital converter (ADC) configured to digitize a plurality of input signals, wherein each input signal includes radar chirp and radar chirp reflection information received at one of a plurality of receiver antennas. The radar data processing device also includes Fast Fourier Transform (FFT) logic configured to generate FFT output samples based on each digitized input signal, wherein at least some of the generated FFT output samples are across antenna FFT output samples associated with at least two of the plurality of receiver antennas. The radar data processing device also includes a processor configured to determine a plurality of object parameters based on at least some of the generated FFT output samples, wherein the processor uses a neural network classifier trained to provide a confidence metric for at least one of the plurality of object parameters.
1. A radar data processing device comprising:
Fast Fourier Transform (FFT) logic configured to generate FFT output samples; and
a processor configured to:
determine, using a neural network classifier, a confidence metric based on the FFT output samples;
adjust a threshold based on the confidence metric; and
detect an object based on the FFT output samples and further based on the adjusted threshold.
2. The radar data processing device of claim 1 ,
wherein the FFT output samples include across antenna FFT output samples, and
wherein the processor is configured to determine, using the neural network classifier, the confidence metric based on the across antenna FFT output samples.
3. The radar data processing device of claim 2 ,
wherein the processor is further configured to analyze coherency information for the across antenna FFT output samples, and
wherein the processor is configured to determine, using the neural network classifier, the confidence metric based on the coherency information for the across antenna FFT output samples.
4. The radar data processing device of claim 2 ,
wherein the processor is further configured to perform a signal strength cleanliness analysis of the across antenna FFT output samples, and
wherein the processor is configured to determine, using the neural network classifier, the confidence metric based on the signal strength cleanliness analysis of the across antenna FFT output samples.
5. The radar data processing device of claim 1 ,
wherein the detected object is a first object, and
wherein the processor is configured to:
determine how signal strength is scattered or concentrated around a number of elevation-azimuth bins; and
responsive to determining how the signal strength is scattered around the number of elevation-azimuth bins, provide the confidence metric indicating a probability of a second object being present in a corresponding range-Doppler bin.
6. The radar data processing device of claim 1 ,
wherein the detected object is a first object, and
wherein the processor is configured to determine the confidence metric for a range of a second object, for a velocity of the second object, or for a direction of arrival for signals reflected by the second object.
7. The radar data processing device of claim 1 , wherein to determine the confidence metric based on the FFT output samples, the processor is configured to determine the confidence metric based on at least one of:
a strength of signals received by at least two antennas;
a pattern of signal received by the at least two antennas;
a distribution of the FFT output samples as a function of range and Doppler; or
a distribution of the FFT output samples as a function of elevation and azimuth.
8. The radar data processing device of claim 1 , wherein the processor is configured to determine an amount of adjustment for the threshold based on the confidence metric.
9. The radar data processing device of claim 1 , wherein the processor is configured to determine the confidence metric by at least determining whether there is signal strength coherency for range indexes of the FFT output samples.
10. A method comprising:
determining, using a neural network classifier, a confidence metric based on Fast Fourier Transform (FFT) output samples;
adjusting a threshold based on the confidence metric; and
detecting an object based on the FFT output samples and further based on the adjusted threshold.
11. The method of claim 10 ,
wherein the FFT output samples include across antenna FFT output samples, and
wherein determining the confidence metric is based on the across antenna FFT output samples.
12. The method of claim 11 , further comprising analyzing coherency information for the across antenna FFT output samples,
wherein determining the confidence metric is based on the coherency information for the across antenna FFT output samples.
13. The method of claim 11 , further comprising performing a signal strength cleanliness analysis of the across antenna FFT output samples,
wherein determining the confidence metric is based on the signal strength cleanliness analysis of the across antenna FFT output samples.
14. The method of claim 10 ,
wherein the detected object is a first object, and
wherein the method further comprises:
determining how signal strength is scattered or concentrated around a number of elevation-azimuth bins; and
responsive to determining how the signal strength is scattered around the number of elevation-azimuth bins, providing the confidence metric indicating a probability of a second object being present in a corresponding range-Doppler bin.
15. The method of claim 10 ,
wherein the detected object is a first object, and
wherein determining the confidence metric comprises determining the confidence metric for a range of a second object, for a velocity of the second object, or for a direction of arrival for signals reflected by the second object.
16. The method of claim 10 , wherein determining the confidence metric based on the FFT output samples comprises determining the confidence metric based on at least one of:
a strength of signals received by at least two antennas;
a pattern of signal received by the at least two antennas;
a distribution of the FFT output samples as a function of range and Doppler; or
a distribution of the FFT output samples as a function of elevation and azimuth.
17. The method of claim 10 , further comprising determining an amount of adjustment for the threshold based on the confidence metric.
18. The method of claim 10 , wherein determining the confidence metric comprises determining whether there is signal strength coherency for range indexes of the FFT output samples.
19. A non-transitory computer-readable medium having executable instructions stored thereon, configured to be executable by a processor for causing the processor to:
determine, using a neural network classifier, a confidence metric based on Fast Fourier Transform (FFT) output samples;
adjust a threshold based on the confidence metric; and
detect an object based on the FFT output samples and further based on the adjusted threshold.
20. The non-transitory computer-readable medium of claim 19 ,
wherein the FFT output samples include across antenna FFT output samples, and
wherein the instructions to determine the confidence metric include instructions to determine the confidence metric based on the across antenna FFT output samples.
21. The non-transitory computer-readable medium of claim 20 , wherein the instructions are configured to be executable by the processor for further causing the processor to analyze coherency information for the across antenna FFT output samples,
wherein the instructions to determine the confidence metric include instructions to determine the confidence metric based on the coherency information for the across antenna FFT output samples.
22. The non-transitory computer-readable medium of claim 20 ,
wherein the instructions are configured to be executable by the processor for further causing the processor to perform a signal strength cleanliness analysis of the across antenna FFT output samples, and
wherein the instructions to determine the confidence metric include instructions to determine the confidence metric based on the signal strength cleanliness analysis of the across antenna FFT output samples.
23. The non-transitory computer-readable medium of claim 19 ,
wherein the detected object is a first object, and
wherein the instructions are configured to be executable by the processor for further causing the processor to:
determine how signal strength is scattered or concentrated around a number of elevation-azimuth bins;
responsive to determining how the signal strength is scattered around the number of elevation-azimuth bins, provide the confidence metric indicating a probability of a second object being present in a corresponding range-Doppler bin.
24. The non-transitory computer-readable medium of claim 19 , wherein the instructions are configured to be executable by the processor for further causing the processor to determine an amount of adjustment for the threshold based on the confidence metric.