• State TI Thinks Resolved
  • Locked Locked
  • Replies 4 replies
  • Answers 1 answer
  • Subscribers 31 subscribers
  • Views 346 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.

AWR1642: Vehicle Occupant Detection: Question on offline-generated training coefficients

Chris Meng
Mastermind 37726 points
Part Number: AWR1642

Hi,

My customer has got the offline-generated training coefficients of Vehicle Occupant Detection based on ODdemo_usersguide.pdf. But they found the coefficients they got on one board/several boards is not always ok for other boards (same batch) with same test environment. The test results is a little different with same coefficients on different boards. Customer suspected that the little difference on the antenna gain between boards causes the different result. Customer wants to use one set of coefficients for all their boards for occupant detection application as they can't do offline-generated coefficients for each board when production.

Would you pls kindly advise if you have any idea on how to solve this problem?

  • 0
    TI__Guru 69960 points

    Hi Chris,

    Training data may depend on antenna gain and other factors. Let me assign this thread to an expert to provide you a detailed explanation and probable solution.

    Regards,

    Jitendra

  • 0
    TI__Mastermind 21025 points

    Hi Chris,

    Can the customer quantify the differences?  I believe that minor differences are to be expected, not only due to slight antenna manufacturing differences, but slight differences in mounting position, angle, etc.  So I would expect that an occasional frame decision may flip in the wrong direction, especially with a weak target.  But the frame rate decisions should be averaged over several seconds to increase confidence level of the detection.

      -dave

  • 0 in reply to db_woodall
    TI__Mastermind 37726 points

    Dave,

    Let me give an example on the difference.

    Customer got the off-line training coefficients on board A.

    When customer uses this coefficients on board B, some missing detection will happen when people is almost static which will not be mised on board A.

    When customer uses this coefficients on board C, some false alerm may happen.

    The boards are mounted in car in same position/angel when testing.

    If you need more detail info about the difference between different boards with same coefficents, pls kindly let me know. I will try to get from customer.

  • 0 in reply to Chris Meng
    TI__Mastermind 21025 points

    Chris,

    Yes, this is what I mentioned in my post above.  Minor differences are to be expected due to manufacturing and mounting position, which will always be slightly different, even in the same mount.

    The purpose of the coefficients is to "teach" the code what the power in the zone should be when the zone is occupied or empty.  It is reasonable to assume that borderline cases can cause a frame's decision to flip one way or the other. This doesn't mean that the coefficients need to be re-trained for that car - it just means that for whatever reason, the observed power was slightly different.

    Two solutions:

    1) The customer should not rely on each frame decision. Rather, frame rate decisions should be averaged over a few seconds to increase the confidence level of the decision.  This should eliminate the minor differences.

    2)  If the differences survive averaging, they can trying adjusting the Rx Gain in the profileCfg up or down.

      -dave