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: Detecting moving people from a moving platform

Part Number: IWR6843ISK

Hi

We have one of your mmwave dev kits (part # above) and have successfully used it to detect and track moving people from a stationary platform using this firmware:

https://dev.ti.com/tirex/explore/node?node=AErmTL1DcBgtfS0zrdmYHg__VLyFKFf__LATEST

Since we are developing a rover with the aim of person avoidance and would really like to mount the mmwave device on the rover itself. Looking at your information here...

https://dev.ti.com/tirex/explore/node?node=AOVWxEvg-HcESecamNaaLg__VLyFKFf__LATEST

this should be possible and we have done so. We see a point cloud in RViz that does indeed correspond to a person. However when the rover starts to move, and the radar system is no longer stationary with respect to the lab frame, the output may no longer be accurate. Can you confirm whether it is indeed possible to detect people while the radar is moving? And if so, what software should we be using for this?

Many thanks

Dave

  • Hello

    YOu will have to see how the tracker for this exact version of demo response to input point cloud fed to it.

    Detection:

    Detection is based on motion, reflection strengths of the objects in the scene.

    Tracker:

    Please see tracker's working - It uses doppler points to club/group objects and assign tracks to them. 

    You may have to modfy/tune the tracker and point cloud to get the tracks for people be recognized.

    Thank you,

    Vaibhav

  • Thanks for the response. Please could you point to some examples or documentation that might help me get going? For example, you say

    >Please see tracker's working

    Where might I be able to find this?

    You mention detection and suggest that this does not rely on doppler. Does TI provide software to generate a point cloud of the scene without making assignments ie. working out what the radar module is actually looking at. A point cloud in a manner similar to lidar would be fine for our application.

  • Hello,

    The ROS Driver that you referenced in your initial post just displays the detected objects in the scene. Each of these detected objects consist of x,y,z points as well as a radial velocity component. The collection of these detected objects is what defines the point cloud. This collection is then updated at a set frame rate e.g 10 or 20Hz. The point cloud that you are seeing on the ROS Driver Visualization is the standard point cloud that the radar device produces. The nature of radar data is not 1 for 1 with what you would expect for LIDAR data.

    If you are interested in explore the performance of how our tracker algorithm performs then I would suggest running the Area Scanner Lab located in our mmWave Industrial Toolbox. Here is the user guide to run this lab:

    Area Scanner User Guide:
    <MMWAVE_INDUSTRIAL_TOOLBOX_INSTALL_PATH>\labs\area_scanner\68xx_area_scanner\docs\area_scanner_68xx_users_guide.html

    Best regards,

    Connor Desmond

  • Thanks for this Connor. To clarify, I'm happy with the performance of the radar module when it is mounted on a stationary platform however, I'd like to get a point cloud from a moving rover. Is this possible? At the moment, it looks like the point cloud disappears when the rover is moving ie. moving the radar module has some effect on the measurements and my point cloud is no longer available. This is the central question to this post.

  • Hello,

    As mentioned in my post above the Area Scanner above, which contains the tracker, is what I would suggest starting out with. From my experience the configuration file for that lab for most use cases works well for when the radar is mobile. Once you have played around with that a bit. You can then take that configuration to the ROS environment to continue evaluation.

    Best,

    Connor

  • Thanks. We'll give that a try.