CFAR adaptive processing for real-time prioritization
Disclosed herein are systems and methods for adaptive CFAR detection and prioritization of objects in an environment. The system can select a cell under test (CUT) and an associated feature set of the cell. A CFAR detection threshold may be determined based on the feature set. The CUT is compared with its neighboring cells to determine whether an object is detected based on the CFAR detection threshold. The detected object is grouped by feature set and compared to a figure of merit (FOM) threshold to generate a priority score. Scored detected objects are sorted to generate a prioritized list of detected objects. The CFAR detection threshold and/or FOM threshold is dynamically updated based on the prioritized list of detected objects.
1 . A radar system for performing object detection on a cell under test (CUT) associated with range-Doppler radar values, comprising:
a computing device processor; and
a memory device including instructions that, when executed by the computing device processor, enables the radar system to:
determine a feature set for the cell under test, the cell under test associated with a plurality of cells, the feature set specifying parameter data including at least one of range data, doppler data, signal-to-noise ratio (SNR) data, noise data, or power data;
select a baseline detection threshold based on the feature set;
analyze parameter data associated with the cell under test with respective parameter data associated with neighboring cells to generate detected unit data indicative of an object detection;
compare the detected unit data to the baseline detection threshold to determine whether the cell under test corresponds to a valid object;
compare the detected unit data to a figure of merit comprising a threshold derived from statistical aggregation of prior radar scan outputs to generate a priority score indicative of detection confidence;
assign the priority score to the detected unit data;
sort the detected unit data based on the priority score to generate a sorted list of detected unit data, wherein the sorted list of detected unit data is partitioned into a plurality of buckets defined jointly by the feature set and the figure of merit;
prioritize the detected unit data using a scheduler that weights factors comprising an order of processing buckets defined by at least one of the feature set or the figure of merit, a weighted distribution among the buckets, and a volume of detected unit data within each bucket to be processed;
update the baseline detection threshold based on the sorted list, wherein the update is performed based on distribution of detected unit data across the plurality of buckets and system-level processing constraints; and
use the feature set associated with the sorted list of detected unit data for further processing.
2 . The radar system of claim 1 , wherein the figure of merit is updated based on radar system constraints including environment considerations, and wherein the figure of merit is jointly tuned with the baseline detection threshold based on statistical aggregation of prior radar scan outputs.
3 . The radar system of claim 2 , wherein the radar system constraints include a processing capacity constraint defining a maximum number of detected unit data items per radar scan cycle.
4 . The radar system of claim 3 , wherein the processing capacity constraint is used by the scheduler to dynamically adjust a volume of detected unit data assigned to each bucket in the plurality of buckets.
5 . The radar system of claim 1 , wherein the scheduler further comprises a priority tracker configured to maintain a hierarchy of the plurality of buckets, and to determine a volume of detected unit data within each bucket in the plurality of buckets based on the hierarchy.
6 . The radar system of claim 5 wherein the priority tracker further comprises a counter configured to track a number of detected unit data items from the each bucket that are added to a prioritized processing queue.
7 . The radar system of claim 1 wherein the instructions, when executed by the computing device processor, further enables the radar system to:
use a machine learning based approach to determine the baseline detection threshold for the feature set and the figure of merit based on prior radar scans and detected object distribution statistics.
8 . A computer-implemented method for performing object detection using a radar system on a cell under test associated with range-Doppler radar values, comprising:
obtaining radar data for performing object detection on a cell under test (CUT), the radar data associated with range-Doppler radar values;
determining a feature set for the cell under test, associated with multiple cells, the feature set specifying parameter data including at least one of range data, doppler data, SNR data, noise data, or power data;
selecting a baseline detection threshold based on the feature set;
analyzing parameter data of the cell under test and neighboring cells to generate detected unit data indicating object detection;
detecting an object by comparing the detected unit data to the baseline detection threshold to determine whether the cell under test corresponds to a valid object detection outcome;
generating a priority score by comparing the detected unit data to a figure of merit comprising a threshold derived from statistical aggregation of prior radar scan outputs, the priority score being indicative of detection confidence;
assigning the priority score to the detected unit data;
sorting the detected unit data based on the priority score to form a sorted list, wherein the sorted list is partitioned into a plurality of buckets defined jointly by the feature set and the figure of merit;
prioritizing the detected unit data using a scheduler that weights factors comprising an order of processing buckets defined by the feature set and the figure of merit, a weighted distribution among the buckets, and a volume of detected unit data within each bucket to be processed; and
updating the baseline detection threshold using the sorted list based on a distribution of detected unit data across the plurality of buckets and system-level processing constraints.
9 . The computer-implemented method of claim 8 , further comprising:
updating the figure of merit based on radar system constraints including environment considerations, wherein the figure of merit is jointly tuned with the baseline detection threshold based on statistical aggregation of prior radar scan outputs.
10 . The computer-implemented method of claim 9 , wherein the radar system constraints include a processing capacity constraint defining a maximum number of detected unit data items per radar scan cycle.
11 . The computer-implemented method of claim 8 , further comprising:
updating one of the baseline detection threshold or the figure of merit to satisfy a system performance metric, wherein the system performance metric is associated with distribution of detected unit data across the plurality of buckets.
12 . The computer-implemented method of claim 11 , wherein the system performance metric includes an amount of detected objects.
13 . The computer-implemented method of claim 8 , further comprising:
processing detected unit data satisfying a priority score threshold determined dynamically based on distribution of detected unit data across the plurality of buckets and system-level processing constraints.
14 . The computer-implemented method of claim 8 , further comprising:
using a machine learning based approach to determine the baseline detection threshold for the feature set and the figure of merit based on prior radar scan distributions.
15 . A non-transitory computer readable storage medium storing instructions that, when executed by at least one processor of a computing system, causes the computing system to:
obtain radar data for performing object detection on a cell under test (CUT), the radar data associated with range-Doppler radar values;
determine a feature set for the cell under test, associated with multiple cells, the feature set specifying parameter data including at least one of range data, doppler data, SNR data, noise data, or power data;
select a baseline detection threshold based on the feature set;
analyze parameter data of the cell under test and the neighboring cells to generate detected unit data indicating object detection;
detect an object by comparing the detected unit data to the baseline detection threshold;
generate a priority score by comparing the detected unit data to a figure of merit, the priority score being indicative of detection confidence;
assign the priority score to the detected unit data;
sort the detected unit data based on the priority score to form a sorted list, wherein the sorted list is partitioned into a plurality of buckets defined jointly by the feature set and the figure of merit;
prioritize the detected unit data using a scheduler that weights factors comprising an order of processing buckets defined by at least one of the feature set or the figure of merit, a weighted distribution among the buckets, and a volume of detected unit data within each bucket to be processed,
wherein the scheduler further comprises a priority tracker configured to maintain a hierarchy of the plurality of buckets and to determine the volume of detected unit data within each bucket to be processed based on the hierarchy; and
update the detection threshold using the sorted list of detected unit data.
16 . The non-transitory computer readable storage medium of claim 15 , wherein the instructions, when executed by the at least one processor, further enables the computing system to:
update the figure of merit based on radar system constraints including environment considerations, and wherein the figure of merit is jointly tuned with the baseline detection threshold based on statistical aggregation of prior radar scan outputs.
17 . The non-transitory computer readable storage medium of claim 16 , wherein the radar system constraints include a processing capacity constraint defining a maximum number of detected unit data items per radar scan cycle.
18 . The non-transitory computer readable storage medium of claim 15 , wherein the priority tracker further comprises a counter configured to track a number of detected unit data items from the each bucket that are added to a prioritized processing queue.
19 . The non-transitory computer readable storage medium of claim 18 , wherein the instructions, when executed by the at least one processor, further enables the computing system to dynamically update the baseline detection threshold and the figure of merit based on a system performance metric.
20 . The non-transitory computer readable storage medium of claim 15 , wherein the instructions, when executed by the at least one processor, further enables the computing system to:
process detected unit data satisfying a priority score threshold.