AWRL6844 CPD Users Guide

Table of Contents

Overview

This lab demonstrates the use of TI 60GHz mmWave sensors AWRL6844 for in-cabin sensing applications including Vehicle Intruder Detection (ID), Child Presence Detection (CPD) and Seat Belt Reminder (SBR). CPD and SBR will share the same signal processing chain for point cloud generation. Note that this lab is tested and tuned with AWRL6844EVM board and top front console mounting. Optimization effort will be needed to achieve best performance for other antenna design and other mounting position.

The same demo binary and visualizer will be used for all three applications, i.e., ID, CPD and SBR. Different configuration files can be loaded which enable the three different applications.

In this users guide, we will focus on the CPD application. Real-time demo recording clips are shown below to demonstrate the CPD performance in different test scenerios. The accumulated point clouds display across 30 frames. The blue box indicates a child is detected and the green box indicates an adult is detected. Currently, the classifier is per-seat based. A per-car decision can be added at visualizer level. Since we are moving the baby into different position with demo running continuously, users may observe some transition period in the probability figure in this video. There are some consistent false points around the car, which are the strong static reflections leaking to the non-zero Doppler bins and being picked up by the super sensitive CFAR. Usually, they are not significent enough to cause a false detection in the classifier.

Requirements

The same hardware and software requirement between CPD and SBR mode. Users can refer to SBR user’s guide for details.

Quickstart

The same visualizer is shared between CPD and SBR mode. Users can refer to SBR User’s Guide for details and make the following modification to the procedure:

Load the cpd.cfg as an example CPD configuration file when starting the demo visualizer.

Note
If performance needs to be improved please review configuration settings. Every car is different and may need modifications to the configuration file to improve performance. See the SBR User’s Guide Section “EVM Mounting and Coordinate Transforms” for detailed description on properly measuring the car for the right configuration settings. See the note on zOffset tuning before for additional calibration that can be done.

Algorithm Overview

This demo can be viewed as two pieces: the point cloud detection and the high-level processing.

• The point cloud detection is the same between CPD and SBR mode. Users can refer to the SBR User’s Guide for details.
• High-level processing is different between CPD and SBR mode. For CPD mode, first ML-based classifier is used to make a decision whether a seat zone is occupied or not. If it is occupied, another ML-based classifier is used to make a decision whether a seat zone is occupied by a child or adult.

The EVM Mounting, Antenna Configuration and Zone definition are also the same between CPD and SBR mode. Users can refer to SBR User’s Guide for details.

Demo Configuration File Parameters are also the same between CPD and SBR mode. We are listing the different default setting here.

Command	Referance
2.	CLI commands that configures the point cloud chain
dynamicRACfarCfg	refer to AWR6843 CPD with Classification, Section CFAR Detection
	range bin start index: set to 5
	range bin end skip: set to 10
	angle start index: set to 1
	angle end skip: set to 1
	search window size - “far” range, set to 4
	search window size - angle dimension, set to 2
	search window size - “near” range, set to 4
	guard window size - “far” range, set to 16
	guard window size - angle, set to 3
	guard window size - “near” range, set to 16
	“K0” cross range detection threshold, set to 6.0
	“K0” cross angle detection threshold, set to 2.0
	“sidelobe” threshold used in the second search pass, set to 0.5
	second search pass enable flag: set to 1
	RefRangeBinIdx: set to 1 to disable dynamic CFAR threshold in range dimension.
	Note that the CFAR thresholds are set low to help the low SNR cases like 2nd-row rear-facing baby and footwell area behind the seats. Guard window and search window size for “near” are set to be the same as “far” to disable the “near” option.
dynamic2DAngleCfg	zoom-in Factor for finer angle estimation: set to 5
	Number of coarse neighboring angle bins of zoom-in. Only 1 is supported
	Number of samples on each side for peak expansion in zoom-in angle heatmap, set to 1.
	peakExpRelThr: peak expansion sharpness threshold, set to 0.1 to encourage peak expansion.
	peakExpSNRThr: peak expansion SNR threshold, set to 1.0, super low to encourage peak expansion.
	localMaxCheck: set to 2 to only allows local peak
runningMode	set to 2 for CPD
featExtrCfg	maxNumPtsPerZonePerFrm: set to 200
	numFrmProc: set to 30 to accumulate the point clouds from the most recent 30 frames for classification
	offsCorr: set to 1
	dbScanFilt: set to 0 to disable dBScan clustering
	dbScanEps: does not matter due to dBScan clustering being disabled
	dbScanMinPts: does not matter due to dBScan clustering being disabled
	Note that featExtrCfg settings should match the ones used for classifier model training.
zOffset	per zone height offset. This is important for CPD tuning. The zOffset is given in meters and is subtracted for all point cloud detections in the z-axis for a given zone. This can be used for calibrating for child vs. adult classification by observing the measured height and compensating. For example, if an adult is detected too short in a given zone, then the zOffset should be negative for that zone to compensate. Similarly, if a baby or child is detected to be too tall, then the zOffset should be positive for that zone to compensate. TI has found that height accuracy is important to child and adult classification accuracy. By calibrating the zOffset, the height estimation accuracy in each zone can be modified to improve classification results.

Tuning

The key items to focus on for tuning are as follows:

• Proper mounting
  The LPD/CPD classifer model has been trained with the sensor mounted at the front overhead console position.
       % Sensor position and tilt angles
     % <xOffset> <yOffset> <zOffset> <azimuthTilt> <elevationTilt>
     sensorPosition 0 0.8 1.15 0 -60
  The performance will be similar to our findings when the sensor being mounted is in a similar position and a similar tilting angle even in different cars. For other positions like center mount, side mount, and back mount, the performance has not been tested at this time..

• Accurate zone definition
  Follow the same procedure in SBR user guide to match the zone definition with car interior spacing.

• The LPD and CPD Classifier
  The classifiers are trained with the following feature extraction parameters, and modification of these paramters can result in non-optimal solutions.
      % SBR feature extraction parameters
    featExtrCfg 200 30 1 0 0.4 10

• Point Cloud Tuning
  Point cloud quality is important to the classifier. Users can adjust CFAR detection threshold “K0 cross range” and “K0 cross angle” in dynamicRACfarCfg to achieve richer point cloud. Note that Lower CFAR detection threshold “K0” results in richer point cloud but higher false detection ratio. Users can also lower “Linear SNR threshold” in dynamic2DAngleCfg to lower the point cloud extension threshold to include more neighboring points next to the detected peak.

• Additional zone adjustment
  Note that for CPD, we have lower the CFAR setting to help on the 2nd-row rear-facing baby and footwell area behind the seat. This change makes the detection more sensitive (compare to cfar in SBR setting) to false detection from the static leakage. If consistent false points is observed from certain area such as headrest and storage area, and these points contribute to the wrong classifier output, customer can consider adjusting the zone definition to remove those areas.

Known Issues and Limitations

Need More Help?

• Search for your issue or post a new question on the mmWave E2E forums
• See the SDK for more documentation on various algorithms used in this demo. Start at <MMWAVE_LSDK6_INSTALL_DIR>\docs\mmwave_sdk_module_documentation.html