• State
  • Locked Locked
  • Replies 8 replies
  • Subscribers 30 subscribers
  • Views 658 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.

IWR6843ISK-ODS: Difference between distance measured by radar and ground truth

Pushkar Mahajan
Prodigy 150 points
Part Number: IWR6843ISK-ODS

Tool/software:

Hello,

I have IWR6843ISK-ODS and using 3D people detection binary file with ros2 to get the point cloud data. I am using this radar with a time-of-flight camera sensor inside a vehicle to measure the distance of the driver from the sensor. Both the radar and the camera are placed very close together and mounted inside the vehicle, aiming at the driver. The distance between the driver and the sensors is in the range of 0.5-1.0 meters.

I observed that the distance measured by the time of flight camera for a joint location (lets say right shoulder) was accurate with physical measurements. However, the distance from radar is always greater than the physical measurement and the one from time of flight camera. The difference is around 8-10 cm between the two, radar giving higher distance value. This is same for other joint locations. 

Could you please explain the reason for higher value of distance from radar? And how to correct it?

Thank you.

Best regards,
Pushkar Mahajan

  • 0
    TI__Mastermind 23107 points

    Hello Pushkar,

    This sounds like your detected points are off by one-two range bins. I do not exactly know what your chirp configurations' range resolution is and at close range (<1m) radar can struggle with positional accuracy, but 10cm off is a little bit higher than expected. Do you see this 8-10cm offset at longer distances? For example 2m to 3m away?

    Best Regards,

    Pedrhom

  • 0 in reply to Pedrhom Nafisi
    Prodigy 150 points

    Hi Pedrhom,

    With the current chirp configuration, my range resolution is 6.3 cm. 

    How much offset is expected for measruements within 1m distance? 

    I observe similar offset at 2 meters as well. 

    Could you explain what could be the reason for this and how to correct it?

    Thank you and best regards,

    Pushkar

  • 0 in reply to Pushkar Mahajan
    Prodigy 150 points

    Hi Pedrhom,

    Its been a while and would like to know your response on it. 

    Best regards,
    Pushkar

  • 0 in reply to Pushkar Mahajan
    TI__Mastermind 23107 points

    Hello Pushkar,

    At the moment I can at least assure you what you are seeing is not typical in terms of consistent range bin discrepancy regardless of distance of object.I know you are using ODS, so my next question is are you using the overhead chain (ceiling mount/ CM) or the standard wall mount chain (WM)? Have you tried the other?

    Best Regards,

    Pedrhom

  • 0 in reply to Pedrhom Nafisi
    Prodigy 150 points

    Hi Pedrhom,

    I am using the standard wall mount chain. 

    I tried the overhead chain with the standard cfg parameters listed in 3D people detection document, but it didn't work. 

    Best regards,
    Pushkar

  • 0 in reply to Pushkar Mahajan
    TI__Mastermind 23107 points

    Hello Pushkar,

    The overhead chain changes a lot of the processing timing needed for a config to start. Also some of the parameters, although named identically, will have fields that completely change what the number represents such as dynamicRangeAngleCfg and dynamic2DAngleCfg when set to overhead mode. Did you try using a default overhead config? You can also try flashing the overhead people tracking binary, the only difference between this binary and the one you are currently using is some memory optimizations for maximizing radar cube size to allow the most high end of configurations with the largest amounts of processing, but functionally is identical.

    Best Regards,

    Pedrohm

  • 0 in reply to Pedrhom Nafisi
    Prodigy 150 points

    Hi Pedrhom,

    Thank you for the information.

    1. In the document "Detection Layer Parameter Tuning Guide for the 3D People Tracking Demo" I didnt find the differences you mentioned about the parameters. Could you direct me to the documentation where this information is available?

    2. In ros2 package, there is no binary file with overhead people tracking as you mentioned. From where can I get it?

    3. I tried the overhead detection with the parameter values listed in the "Detection Layer Parameter Tuning Guide for the 3D People Tracking Demo" and it didint work. Is that the default overhead config?

    4. I performed the data collection with wall mounting configuration and would like to know the possible explanation for 8-10 cm difference to mention. i understand that there is 1or 2 range bin shift that could result in this difference but how to explain why it happened?

    Thank you.

    Best regards,
    Pushkar

  • 0 in reply to Pushkar Mahajan
    TI__Mastermind 23107 points

    Hello,

    1. In that same document, compare pages 29 and 31 for example. Page 29 dynamic2DAngleCfg shows what the fields are when rangeAngleEstMethod = 1 which is standard People Tracking wall mount (3DWM). Page 31 is when rangeAngleEstMethod = 2 which is overhead 3D People tracking when it is ceiling mounted (3DCM).

    2. You can get it from the Toolbox on dev.ti.com. Get this specific version of the Industrial Toolbox as I am certain it will work with the ROS Driver: https://dev.ti.com/tirex/explore/node?a=VLyFKFf__4.12.0&node=A__AEIJm0rwIeU.2P1OBWwlaA__com.ti.mmwave_industrial_toolbox__VLyFKFf__4.12.0

    It will be under the old name overhead 3d people counting C:\ti\mmwave_industrial_toolbox_4_12_0\labs\People_Counting\overhead_3d_people_counting\prebuilt_binaries\default

    3. You can find a config in a folder within that labs directory as well.

    I am not currently sure what could be causing your range bins to be consistently off. Do you see the same behavior with the out of box demo binary?

    Best Regards,

    Pedrhom