• State
  • Replies 5 replies
  • Subscribers 33 subscribers
  • Views 140 views
  • Users 0 members are here
Support feedback
Options
Options
Related

IWR6843: [CLI CONFIG] Help me config CLI RADAR for my Application

Nguyen Trung
Prodigy 10 points

Part Number: IWR6843

We are currently developing a ceiling-mounted multi-zone human presence detection sensor.

The sensor is designed to:

  • Track up to five people simultaneously.
  • Divide the monitored area into multiple independent detection zones.
  • Continuously track each person's position.
  • Maintain reliable presence detection even when a person remains almost motionless, such as while working at a desk or sleeping.

The functionality is similar to the Aqara FP2, but our product is optimized for ceiling-mounted installation at a height of 2 to 4 meters.

We would appreciate your support in recommending the optimal radar configuration and parameter settings for this application.

The configuration we are currently using is provided below. We would be grateful for any suggestions or optimizations you can recommend. pt_6843_3d_aop_overhead_customdemotest.cfg 

  • 0
    TI__Intellectual 1631 points

    Hi Nguyen,

    Can you please tell me how the current configuration is not optimal? We can't make suggestions to improve its performance if we don't know what it isn't doing that you would like it to.

    Best,

    David

  • 0 in reply to David Cassity
    Prodigy 10 points

    Hello David Cassity,

    (Image 1) 

    Additional Test Information

    • Device: IWR6843AOPEVM
    • Demo: Overhead 3D People Tracking
    • Mounting Height: 2.2 m
    • Ceiling Mount: Yes
    • Person walking toward radar from different directions.
    • East-West initial Track ID distance: ~3 m
    • North-South initial Track ID distance: ~2 m
    • Point Cloud density in North-South is visibly lower than East-West.

    We are currently evaluating the IWR6843AOPEVM using the Overhead 3D People Tracking Demo as the baseline for developing our own ceiling-mounted presence sensor.

    During our testing, we observed that the initial detection distance (when the first Track ID is created) is not consistent across different directions.

    As shown in the attached figure: (Image 2, image 3)

    • The radar is mounted on the ceiling at approximately 2.2 meters.
    • Along the East-West (E-W) direction, a person can be detected and a Track ID is created at approximately 3 meters from the radar.
    • However, along the North-South (N-S) direction, the first Track ID is not created until the person is much closer, at approximately 2 meters.

    This means the effective detection range is significantly different depending on the direction.

    In addition, we also observed that:

    • The Point Cloud density in the North-South directions is noticeably lower than in the East-West directions.
    • Because of the lower Point Cloud density, the tracker creates Track IDs much later in these directions.
    • As a result, the detection coverage is not uniform around the radar.

    For a ceiling-mounted presence sensor, we expect the detection range to be approximately symmetric in all directions, especially since the advertised FoV is around ±60°. However, the practical detection range in the North-South directions appears to be much smaller than expected.

    We would like to ask the following questions:

    1. Is this behavior expected for the Overhead 3D People Tracking Demo?
    2. Is the reduced Point Cloud density in the North-South directions caused by the antenna pattern of the IWR6843AOP, or by the signal processing algorithms?
    3. Are there any recommended parameters in the Detection Layer or Tracking Layer that can improve the Point Cloud density and early detection performance in these directions?
    4. Would you recommend modifying profileCfg, dynamicRACfarCfg, dynamicRangeAngleCfg, allocationParam, or any other parameters to achieve a more uniform detection range?
    5. Have you observed similar behavior in other ceiling-mounted presence sensor applications?

    Our target application is a commercial ceiling-mounted multi-zone presence sensor that needs to:

    • Detect people as early as possible.
    • Maintain reliable presence detection when people are standing, sitting, working, or sleeping.
    • Support up to five people simultaneously.
    • Provide a uniform detection range in all directions.

    Any recommendations or best practices would be greatly appreciated.

    Image:

    drive.google.com/.../1WbOL4N2NPOexQv2OEyX7VzvEQAVmc5fj

  • 0 in reply to David Cassity
    Prodigy 10 points

     Please support me this case!

  • 0 in reply to Nguyen Trung
    TI__Mastermind 29172 points

    Hello,

    Sorry for the delay here. Can you provide any updates to the issues you've seen? As well as what issues/concerns are your highest priority

    Best Regards,

    Pedrhom

  • 0 in reply to Pedrhom Nafisi
    Prodigy 10 points
    Hi Pedrhom,  

    I am currently testing the wall-mounted version of the algorithm. However, I am encountering issues with multi-person tracking.

    Case 1:  A person is sitting still while another person walks past nearby.
    I tested the following tracking configuration: (wall)
    gatingParam 3 2 2 3 4
    However, I still observed an issue. When one person is sitting completely still and another person walks nearby, the track of the stationary person is sometimes pulled toward the moving person and eventually lost.
    A new track for the stationary person is only created after they make a significant movement. If the person remains still, no new track is generated.
    For our presence sensing application, this is a critical issue because the light may turn off even though the person is still present in the monitored area.
    Is there any tracking parameter that can reduce track merging or prevent a stationary person's track from being associated with a nearby moving person?
    Case 2: Stable multi-person tracking.
    Position: 1,7 meters
    For the wall-mounted version, I am having difficulty handling scenarios with multiple people.
    I created two monitoring zones around the desks, but I observed that the lights sometimes turn off when another person walks nearby. In some cases, the stationary person's track is lost.
    Is there any way to optimize the tracking algorithm for multi-person scenarios (up to 3–5 people)? Our priority is to maintain reliable presence detection for people who are sitting still, even when others are moving around them. We need to avoid track switching, track merging, or track loss, as these issues can cause the lights to turn off while someone is still present.
    Is there a better solution for multi-person tracking?