• State TI Thinks Resolved
  • Locked Locked
  • Replies 11 replies
  • Answers 1 answer
  • Subscribers 31 subscribers
  • Views 844 views
  • Users 0 members are here
Support feedback
Options
Options
Related

This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

IWR6843: Points detected with no movement

Greg Yamanishi48
Expert 1130 points
Part Number: IWR6843
Other Parts Discussed in Thread: MMWAVEICBOOST,

I am using a MMWAVEICBOOST board + IWR6843.

I am running lab 15.  Peoplecount demo for 68xx.  I see points detected all over the field, even though nothing is moving.  Compared to the 1642 it is not working as well.

Tracking is less accurate, and targets appear when they are not there.  Can you recommend some parameter changes?

  • 0
    TI__Mastermind 28055 points
    Hi Greg,

    Newest release has a configuration change that fixes a bug. I have posted the configuration fix here for your convenience:
    flushCfg
    dfeDataOutputMode 1
    channelCfg 15 5 0
    adcCfg 2 1
    adcbufCfg 0 1 1 1
    profileCfg 0 60.6 30 10 62 0 0 53 1 128 2500 0 0 30
    chirpCfg 0 0 0 0 0 0 0 1
    chirpCfg 1 1 0 0 0 0 0 4
    frameCfg 0 1 128 0 50 1 0
    lowPower 0 1
    guiMonitor 1 1 0 0
    cfarCfg 6 4 4 4 4 16 16 4 4 50 62 0
    doaCfg 600 1875 30 1 1 0
    SceneryParam -6 6 0.5 6
    GatingParam 4 3 2 0
    StateParam 10 5 100 100 5
    AllocationParam 250 250 0.25 10 1 2
    AccelerationParam 1 1 1
    trackingCfg 1 2 250 20 52 82 50 90
    sensorStart

    Let me know if this helps.

    Regards,
    Justin
  • 0 in reply to Justin Curewitz
    Expert 1130 points
    I compared this cfg file with the one I am using and they are identical.
  • 0 in reply to Greg Yamanishi48
    TI__Mastermind 28055 points

    Hi Greg,

    If possible, can you post pictures or video that describe the issue you are seeing?  Right not I can recommend increasing this bolded value:

    AllocationParam 250 250 0.25 10 1 2

    You should increase this until you no longer have ghost detections.  This value sets the number of points required to allocate a track in the tracker.

    Regards,

    Justin

  • 0 in reply to Greg Yamanishi48
    Expert 1130 points
    Attached is a zipped mp4 file of the screen capture. Using the default cfg file (like the one you mention above). Issues:
    1. When there is no movement in the zone there are a constant set of points being gathered with no objects moving. This was the problem logged. You can see dots appearing up to 2 meters away. There is nothing moving in the area.
    2. When a person walks through the zone, I am getting a ghost image. Later on in the video, a person walks under and past the sensor.
  • 0 in reply to Greg Yamanishi48
    TI__Mastermind 28055 points

    Hi Greg,

    I will need another day to look at this.  In the meantime, you can try changing the CFAR thresholds, to see if it is possible to remove the false detections that way. Find this line in the config:

    cfarCfg 6 4 4 4 4 16 16 4 4 50 62 0

    And try increasing the bolded parameters in increments of 5. If you can eliminate most of the false detection points this way, please test to see if this can still detect a walking person.

    Finally, could you post any changes you have made to the configuration?

    Thank you,

    Justin

  • 0 in reply to Justin Curewitz
    Expert 1130 points

    The visualization is showing noise points within 1m and 2m and 3m.

    cfarCfg 6 4 4 4 4 16 16 4 4 50 62 0

    running with the default config, the targets did not appear till 3m from the sensor.  I was not getting multiple targets today for 1 person

    cfarCfg 6 4 4 4 4 16 16 4 4 55 67 0

    cfarCfg 6 4 4 4 4 16 16 4 4 60 22 0

    These configurations picked up targets quicker, but not consistently.  Sometimes targets were not detected

    cfarCfg 6 4 4 4 4 16 16 4 4 65 77 0

    Targets were not created

    These data points are very anecdotal.  It is hard to get repeatable results.  I suffices to say that the visual representation was difficult to identify if a person was being tracked.  

    1. one person moved and stopped.  Another person moved and stopped.  Only one target created.

    2. one person moves and goes behind the sensor.  The target stays and never goes away.  (this could be a matlab bug).

    3. Sometimes the centroid gets very large.  I assume the point cloud is spread out.  Is that the case.

    I tried multiple people, in none of these configurations did two targets get created.  At one point two targets appeared and disappeared.  Generally it was identified as one person.

  • 0 in reply to Greg Yamanishi48
    Expert 1130 points
    AllocationParam 500 250 0.25 50 1 2

    This was the allocation adjustments.
  • 0 in reply to Greg Yamanishi48
    TI__Mastermind 28055 points
    Hi Greg,

    The performance you are describing does not seem to be the behavior we would normally expect, even for a People Counting demo with poor performance. I request that

    1) Please capture and send video of the gui with the default configuration. I can compare this to the performance my device gives with the default config. If possible, please collect a video of the person walking that corresponds to the gui output.

    2) Can you send a picture of your device?

    Thanks,
    Justin
  • 0 in reply to Justin Curewitz
    Expert 1130 points
    Please see my post from Feb 1. It has a mp4 of the screen capture from the Matlab demo
  • 0 in reply to Greg Yamanishi48
    TI__Mastermind 28055 points

    Hi Greg,

    Can you run the OOB demo and capture the following?

    1. Range Doppler Heatmap with static clutter removal off
    2. Range Doppler Heatmap with static clutter removal on
    3. Range-Azimuth Heatmap with static clutter removal off
    4. Range-Azimuth Heatmap with static clutter removal on

    For these captures, please set the device in an area without any significant clutter - empty if possible.  Please use the default configuration in the OOB Demo Visualizer, "Best Range Resolution".  Please take screenshots of the Heatmap output.  We are expecting to see strong signal strength where there is nothing - especially in the capture with static clutter removal on, if there are no moving objects in the scene, there should be no areas of high signal strength.

    You can also try turning down the gain in the default config: profileCfg 0 60.6 30 10 62 0 0 53 1 128 2500 0 0 24

    Regards,

    Justin