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: People Count Accuracy improvement in IWR6843 Radar sensor

Part Number: IWR6843

We are using IWR6843 ISK EVM board to count and track people.

Wanted to check how can we improve the People count accuracy as the current observation is around Max 5-6 people are reported correctly.

If there are more people in Radar FoV we have seen variations, like if there are actually 10 people, the Radar Sensor reports 8 people.

  • Hi Rohit,

    This is the performance TI expects out of the current demo.  If you want to experiment with the software, see the customization guide for tuning the tracker, and the document on tracker function for modifying the source code.

    Regards,

    Justin

  • Hi Justin,

    Thanks for your reply.

    We have already tried customization through different parameters.

    This did not help so much in improving the accuracy of people count.

    If possible then let us know practically what is the best possible accuracy to achieve through the IWR6843 EVM board.

    Regards.

    Rohit.

  • Hi Rohit,

    The accuracy numbers you are seeing are in line with the numbers from our TID - see the testing section near the end. What accuracy numbers are you hoping to get? How many people do you need to count?

    Regards,

    Justin

  • Hi Justin,

    We need to count at least 12-15 people accurately.

    But the supported accuracy for 11 people and above is just 6%.

    Regards,

    Rohit.

  • Hi Rohit,

    The IWR6843 won't count 15 people without error with the current software. We have a few challenges to overcome:

    1. High Density of People - due to low angular resolution of the single chip radar device, it is difficult to differentiate people who are close together (about 1 m or less). This is the common cause of missing people.
    2. LOS being blocked - depending on the mounting scenario, line of site can be blocked by people in the scene, making it more difficult for the sensor to detect people behind them.  Mounting the sensor at least 50 cm above head height and tilting the sensor down so that bore-sight falls in the center of the tracking area helps in this situation.

    For further investigation, you may try using multiple IWR6843 sensor, and having all of the send point cloud to one tracker. Having these sensor at different location may help with both problems listed above.

    If you expect to be monitoring a large area (for example a space 50 meters long), the people you are counting may be spread out enough to get high accuracy at larger numbers than listed in the TI Design, but for indoor environments where people are close together, you will still be facing the issues listed above.

    Regards,

    Justin

  • Hi Justin,

    The FoV of 1 Radar is around 700sqft with 8mtr range.

    I am talking about if there are more than 15 people in this FoV moving and say not necessarily standing side by side,

    or having sufficient space in between.

    Then why is the Radar output not able to detect properly.

    Is there any Software update in plan that can resolve this issue to some extent ?

    Regards,

    Rohit.

  • Hi Rohit,

    Each person can create a fairly large point cloud on their own.  When they are nearby eachother, it is difficult for the radar algo to tell if the point cloud is one large cloud or two small clouds.  As more people are added to the scene, it becomes more frequent for this to occur. We are working to improve our tracker, but it is unlikely that singe-chip radar will detect 15 people accurately and consistently in that small area.

    Regards,
    Justin

  • Hi Justin ,

    Thanks for your reply and taking out time to answer my queries.

    So we can conclude that for any use case where 6 or more people are required to be accurately tracked, then the current chip and SW version won't support due to certain limitations.

    Request to let us know if and when the improved radar tracker HW/SW could be available.

    Regards,

    Rohit.

  • Hi Rohit,

    You conclusion is correct. We will continue to update the lab with better versions of the software.

    Regards,

    Justin