• State Resolved
  • Locked Locked
  • Replies 11 replies
  • Subscribers 31 subscribers
  • Views 1085 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.

Original question:

IWR6843ISK: CFG file configuration for long range detection

IWR6843ISK: One track for multiple people in long range people detection

Aleena N A
Intellectual 530 points
Part Number: IWR6843ISK

Hello everyone, 

I am working on long-range people detection IWR6843ISK. With provided cfg there were ghosts. It was somehow rectified by changing the allocation parameters and state parameters. But one track for multiple people is not solved. I have changed the gating parameters and allocation parameters(maxdistancethreshold, maxvelocitythreshold). But it is again showing a single track for multiple people.

How to solve the problem of single track?

Do these parameters affect tracking and ghosts?

Thanks and regards,

Aleena N A

  • 0
    TI__Mastermind 23725 points

    Hello

    Please see the response to similar thread from you.

    (+) IWR6843ISK: CFG file configuration for long range detection - Sensors forum - Sensors - TI E2E support forums

    We may be able to provide your information about the configuration parameters.
    Thank you,
    Vaibhav
  • 0 in reply to Vaibhav Mahimkar
    TI__Genius 14415 points

    Hello Aleena,

    Can you please elaborate on when multiple people are seen as one track? How close together are the people standing? If they are farther apart will they be tracked separately? At what range(s) are you seeing these issues? The gating parameters are the main params to change the allocation of unique objects, but also the long range profiles have a larger range resolution, so you likely get fewer distinct points per person than other profiles.

    Regards,

    Jackson

  • 0 in reply to Jackson Thomas
    Intellectual 530 points

    Hello Jackson,

    When two people walk at the same time and direction one track is shown. The people will be less than 1 meter apart. If people pass one by one with a delay of a second then it is correctly tracked. For all ranges, the issue is happening. 

    Thanks and regards,

    Aleena N A

  • 0 in reply to Aleena N A
    TI__Genius 14415 points

    Hello Aleena,

    Yes this is what I would expect. It is very difficult to distinguish between two people who are walking together and relatively close. Do you know what the pointcloud looks like when they are walking? Can you see a distinction between the people or does it look like one common shape? Also which profile are you using for these tests? The 50m profile? The angle resolution should be better on the 50m profile than the 100m and might help. 2I think you are already using this cfg though.

    Regards,

    Jackson

  • 0 in reply to Jackson Thomas
    Intellectual 530 points

    Hello Jackson,

    In long rang people detection the point cloud is seen as scattered but in point cloud in people counting demo is seen as clusters for different persons. When two people walk together I can't distinguish the people. I am using 50m 3D profile.

    Thanks and Regards,

    Aleena N A

  • +1 in reply to Aleena N A
    TI__Genius 14415 points

    Hi Aleena,

    This was one of the main reasons we developed the 3D people counting labs. The detection layer uses a different algorithm to generate the point cloud that is more accurate for people. If you are only interested in 50m, there is no difference in the radar chirp generation between regular 3D people counting and the long range version. The algorithm is different, but the main difference in long range is the 100m profile with beamsteering. If you are not using that then 50 meters should be somewhat achievable with the 3D people counting if you set the profileCfg similar to the one in the LRPD lab. This has not been a focus of our testing with that lab so I can't fully comment on performance. But it seems this might give you slightly better results. Have you tried this already?

    In general, distinguishing two people walking closely at the same speed and location is a very difficult task.

    Regards,

    Jackson

  • 0 in reply to Jackson Thomas
    Intellectual 530 points

    Hello Jackson,

    Thank you for the suggestion. I haven't tried this. I will try it.

    Thanks and Regards,

    Aleena N A

  • 0 in reply to Jackson Thomas
    Intellectual 530 points

    Hello Jackson,

    I tried by changing the profileCfg but detecting and tracking were at about 14 m. Is there anything else to change? I would like to know does changing profile CFG affects the range? What defines the range of the demo?

    Thanks and regards,

    Aleena N A

  • +1 in reply to Aleena N A
    TI__Genius 14415 points

    Hi Aleena,

    This comes from some radar physics and theory about the chirp profile. Please see the following document for information about setting the max range.

    Programming Chirp Parameters in TI Radar Devices (Rev. A)

    For tracking you will also need to modify the boundary box and other tracking configuration parameters to allow for further tracking.

    3D_people_counting_tracker_layer_tuning_guide.pdf

    Regards,

    Jackson

  • 0 in reply to Jackson Thomas
    Intellectual 530 points

    Hello Jackson,

    We almost got that. When the target walks from the side of the sensor position, it tracks after a few frames to a distance of 30m. But when the target is coming from the opposite side it is tracked only at a distance of ~15m. 

    Another problem is after a few minutes nothing plots and the exception occurs as 'No plot update' and 'list index out of range'. where the default value corresponding to this can be changed? We have to run it continoulsy.

    Thanks and regards,

    Aleena N A

  • 0 in reply to Aleena N A
    TI__Genius 14415 points

    Hi Aleena,

    Are you using the python visualizer provided? Does it always stop at the same time? Or is it stopping randomly? Sometimes if there is too much open in the background of the PC it can crash.

    When this happens, can you close the Python program and then connect to the device on teraterm or some serial emulator? If you connect to the data port this will show if there is still data coming from the device. This will help us know if the sensor is crashing or if the GUI is crashing.

    Regards,

    Jackson