• State Resolved
  • Locked Locked
  • Replies 9 replies
  • Answers 1 answer
  • Subscribers 30 subscribers
  • Views 1988 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.

IWR1443BOOST: How to increase the number of point cloud data in 3D mode

Katsuhiro Miyamoto
Expert 1460 points
Part Number: IWR1443BOOST

Hello team,

One of my customer's project, the number of point cloud data needs 128 or more in 3D mode. 3D mode, I mean 3TX and 4RX is activated in demo Visualizer for IWR1443BOOST.

Looking at following threads, MMW_MAX_OBJ_OUT in source code of SDK defines the number of detected object but it seems to work only for 2D mode.
https://e2e.ti.com/support/sensor/mmwave_sensors/f/1023/t/647223

https://e2e.ti.com/support/sensor/mmwave_sensors/f/1023/p/624965/2306304

 So could you please explain how to increase the number of pint cloud data in 3D mode? 128 points is minimum they want, if possible 256 to 1028 points is preferable.

Best Regards,
Katsuhiro

  • Former Member
    0 Former Member in reply to Katsuhiro Miyamoto

    Hello Katsuhiro,

    To clarify, it is not accurate to say that  MMW_MAX_OBJ_OUT only works for 2D mode - it does apply in 3D mode.

    In the post you have linked and also for your application, the issues is that increasing MMW_MAX_OBJ_OUT increases the memory allocation to enable the max number of points, but it does not change whether your radar configuration and object of interest will generate that max number of points.

    The post linked is referring to the fact that the point cloud in the azimuth (2D axis) is more rich/has more points because there is better angular resolution in the horizontal axis. In 3D mode, the elevation/vertical axis has less resolution (due to the antenna configuration). Therefore, you will always have better resolution and more points across the horizontal than the vertical axis with the antenna on the 1443.

    So is the problem that you are running out of memory to see the maximum points or your object is not generating enough points?

    Best,

    Amanda

  • 0 in reply to Former Member
    Expert 1460 points

    Hello Amanda,

    My customer observed that the maximum points are 64 in 3D mode, therefore it was probably the case running out of memory.

    Following lines in post_processing.c probably explain that the number of max points in 3D mode is set 64 in default demo software.

    In order to detect and identify the target objects in the customer's project, many points (more than 64) needs to be acquired.
    They would like to get as many number of points as possbile that can be acquired within the internal memory limitation.
    Is there a good solution?

    Best Regards,

    Katsuhiro

  • 0 in reply to Katsuhiro Miyamoto
    Expert 1460 points
    Hello Amanda,

    Do you have any update?

    Best Regards,
    Katsuhiro
  • Former Member
    0 Former Member in reply to Katsuhiro Miyamoto
    Hello Katsuhiro,

    Thank you for clarifying the issue. Given the memory limitation, would enabling persistence of points be a viable solution for your use case? Depending on the update rate you need. You can accumulate points over multiple frames (since the max points is a per frame frame limit) which would enable you to visualize more dense point cloud.

    The Zone Occupancy Detection visualizer, enables this point persistence for 3, 5, 10, and 100 frames.

    dev.ti.com/.../

    Amanda
  • 0 in reply to Former Member
    Expert 1460 points

    Hello Amanda,

    I played with the ZOD visualizer a couple of days. Its 3D visualization is much better than that of demo visualizer and it's easier to use to change config file for testing.

    I modified MMW_MAX_OBJ_ELEVATION to confirm if the number of point cloud output is increased.

    Changed MMM_MAX_OBJ_ELEVATION

     to 100 ⇒ ZOD visualizer frequently hang up.

     to 90  ⇒ ZOD visualizer sometimes hang up.

    To do the test I changed config file as attached to decrease memory allocation (not sure if the config file did so.).

    Can you tell me how to know if memory is full?

    Is it possible to increase the number of point cloud output without affecting memory allocation and processing speed when MMW_MAX_OBJ_ELEVATION and MMW_MAX_OBJ_OUT is increased?

    Best Regards,

    Katsuhiro

    TEST_MMW_MAX_OBJ_ELEVATION_ZOD.xlsx

  • 0 in reply to Katsuhiro Miyamoto
    Expert 1460 points
    The customer said that the enabling persistence of points could not be a good solution in their application.
  • Former Member
    0 Former Member in reply to Katsuhiro Miyamoto

    Hello Katsuhiro,

    It is not possible to increase the point cloud output without affecting memory allocation or processing speed. This is a fundamental trade off in the demo.

    However, please note that  some of the processing limitation is from how the OOB demo is structured and the UART speed. The processing of the point cloud must be complete within each frame.

    If the user is in the algorithm development and evaluation stage perhaps it may be useful for the user to consider using the DCA1000  for raw data capture? With this tool, the user can collect raw data and offline process to enable the most detection possible without concern for memory or speed constraint. Then when the use case is understood, the user can port the algorithm to the device.

    With regards to the ZOD visualizer hang up this can occur if the visualizer cannot keep up with the buffer. A solution would be to try and run the visualizer from the source m file and ensure that the PC is running in high performance mode (power is plugged in). However due to the way the visualizer is implemented the hang up may continue unless code is optimized for plotting a high number of points. (You could enable this by disabling the grid occupancy feature since this is resource intensive for the visualizer).

    Amanda

  • 0 in reply to Former Member
    Expert 1460 points
    Hello Amanda,

    Thank you for detailed explanation. My customer would modify matlab code, take raw ADC data or else.
    I appreciate your support.

    Best regards,
    Katsuhiro