Radar system using a machine-learned model for identifying a number of objects
This document describes techniques and systems related to a radar system using a machine-learned model to identify the number of objects within each range-Doppler bin. For example, the radar system includes a processor that obtains radar data associated with objects and processes the radar data to generate beam vectors. The processor then uses a machine-learned model to identify the number of objects within each range-Doppler bin using extracted features (e.g., magnitude variation, signal-to-noise ratio, subarray beam vector correlations) of the beam vectors. The processor selects a particular angle-finding technique based on whether a single object or multiple objects are identified. In this way, the described systems and techniques more accurately identify the number of object in each range-Doppler bin, thus improving the computational efficiency and robustness of subsequent angle finding.
1. A system comprising:
a radar system including one or more processors configured to identify a number of objects within each range-Doppler bin of a range-Doppler map by:
obtaining radar data associated with one or more objects that reflected a radar signal;
generating, using the radar data, multiple beam vectors, each respective beam vector including the radar data from multiple channels for a particular range-Doppler bin;
extracting, for each respective beam vector of the multiple beam vectors, multiple features of the radar data;
identifying, using a machine-learned model for each respective beam vector of the multiple beam vectors, the number of objects within the particular range-Doppler bin, the machine-learned model being configured to determine whether a single object or multiple objects are present in the particular range-Doppler bin using the multiple features associated with the respective beam vector; and
in response to an output of the machine-learned model classifying the respective range-Doppler bin as having the single object, using a first angle-finding technique to determine an angle associated with the single object for the respective range-Doppler bin; or
in response to the output of the machine-learned model classifying the respective range-Doppler bin as having the multiple objects, using a second angle-finding technique to determine angles associated with the multiple objects for the respective range-Doppler bin, the second angle-finding technique being different than the first angle-finding technique.
2. The system of claim 1 , wherein:
the first angle-finding technique is a fast Fourier transform (FFT) beamforming approach; and
the second angle-finding technique is a super resolution technique.
3. The system of claim 2 , wherein the super resolution technique includes at least one of a multiple signal classification (MUSIC) technique, an estimation of signal parameters via rotational invariance technique (ESPRIT), or an iterative adaptive approach (IAA).
4. The system of claim 1 , wherein the machine-learned model is a logistic regression model with weights associated with each feature of the multiple features and other weights associated with polynomial features of the multiple features.
5. The system of claim 1 , wherein the machine-learned model is a decision tree model that compares a feature value of the respective beam vector to a condition value at each decision.
6. The system of claim 1 , wherein the machine-learned model is a multi-layer perceptron model that uses linear combinations of the multiple features and a non-linear activation function.
7. The system of claim 1 , wherein the multiple features include a magnitude variation, a signal-to-noise ratio, a first subarray beam vector correlation, and a second subarray beam vector correlation of the multiple beam vectors.
8. The system of claim 1 , wherein the multiple features are extracted by the machine-learned model.
9. The system of claim 1 , wherein the machine-learned model is trained offline using ground truth data, the ground truth data including beam vectors for single-object and multiple-object scenarios.
10. The system of claim 1 , wherein the angle is at least one of an elevation angle or an azimuth angle.
11. The system of claim 1 , wherein:
the one or more processors are further configured to output the angle associated with the single object or the angles associated with the multiple objects to an assisted-driving system or autonomous-driving system of a host vehicle; and
the system further comprises the assisted-driving system or autonomous-driving system configured to operate the host vehicle based on the angle associated with the single object or the angles associated with the multiple objects.
12. The system of claim 1 , wherein the system is configured to be integrated in or installed in a vehicle.
13. The system of claim 1 , wherein a type and complexity of the machine-learned model is determined based on a processing capability of the one or more processors of the radar system.
14. A method for identifying a number of objects within each range-Doppler bin of a range-Doppler map, the method comprising:
obtaining radar data associated with one or more objects that reflected a radar signal;
generating, using the radar data, multiple beam vectors, each respective beam vector including the radar data from multiple channels for a particular range-Doppler bin;
extracting, for each respective beam vector of the multiple beam vectors, multiple features of the radar data;
identifying, using a machine-learned model for each respective beam vector of the multiple beam vectors, the number of detected objects within the particular range-Doppler bin, the machine-learned model being configured to determine whether a single object or multiple objects are present in the particular range-Doppler bin using the multiple features associated with the respective beam vector; and
in response to an output of the machine-learned model classifying the respective range-Doppler bin as having the single object, using a first angle-finding technique to determine an angle associated with the single object for the respective range-Doppler bin; or
in response to the output of the machine-learned model classifying the respective range-Doppler bin as having the multiple objects, using a second angle-finding technique to determine angles associated with the multiple objects for the respective range-Doppler bin, the second angle-finding technique being different than the first angle-finding technique.
15. The method of claim 14 , wherein:
the first angle-finding technique is a fast Fourier transform (FFT) beamforming approach; and
the second angle-finding technique is a super resolution technique, the super resolution technique being at least one of a multiple signal classification (MUSIC) technique, an estimation of signal parameters via rotational invariance technique (ESPRIT), or an iterative adaptive approach (IAA).
16. The method of claim 14 , wherein the multiple features include a magnitude variation, a signal-to-noise ratio, a first subarray beam vector correlation, and a second subarray beam vector correlation of the multiple beam vectors.
17. The method of claim 14 , wherein the method further comprises:
outputting the angle associated with the single object or the angles associated with the multiple objects to an assisted-driving system or autonomous-driving system of a host vehicle; and
operating the host vehicle based on the angle associated with the single object or the angles associated with the multiple objects.
18. The method of claim 14 , wherein the multiple features are identified by performing feature engineering on ground truth data, the ground truth data including beam vectors for single-object and multiple-object scenarios.
19. The method of claim 14 , wherein the machine-learned model comprises:
a logistic regression model with weights associated with each feature of the multiple features and other weights associated with polynomial features of the multiple features;
a decision tree model that compares a feature value of the respective beam vector to a condition value at each decision; or
a multi-layer perceptron model that uses linear combinations of the multiple features and a non-linear activation function.
20. A non-transitory computer-readable storage media comprising computer-executable instructions that, when executed, cause a processor of a radar system to identify a number of objects within each range-Doppler bin of a range-Doppler map by:
obtaining radar data associated with one or more objects that reflected a radar signal;
generating, using the radar data, multiple beam vectors, each respective beam vector including the radar data from multiple channels for a particular range-Doppler bin;
extracting, for each respective beam vector of the multiple beam vectors, multiple features of the radar data;
identifying, using a machine-learned model for each respective beam vector of the multiple beam vectors, the number of detected objects within the particular range-Doppler bin, the machine-learned model being configured to determine whether a single object or multiple objects are present in the particular range-Doppler bin using the multiple features associated with the respective beam vector; and
in response to an output of the machine-learned model classifying the respective range-Doppler bin as having the single object, using a first angle-finding technique to determine an angle associated with the single object for the respective range-Doppler bin; or
in response to the output of the machine-learned model classifying the respective range-Doppler bin as having the multiple objects, using a second angle-finding technique to determine angles associated with the multiple objects for the respective range-Doppler bin, the second angle-finding technique being different than the first angle-finding technique.