Because of the holidays, TI E2E™ design support forum responses will be delayed from Dec. 25 through Jan. 2. Thank you for your patience.

  • State Resolved
  • Locked Locked
  • Replies 1 reply
  • Subscribers 30 subscribers
  • Views 617 views
  • Users 0 members are here
Support feedback
Options
Options
Similar topics

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.

Questions on Signal Processing

Yang Tianxi
Intellectual 525 points

in TI's  mmWave device overview_AWR1x.pdf, I found clustering module in the signal processing flow ,

1DFFT->2DFFT-> detection-> 3DFFT(DOA) -> Clustering-> Tracking-> Object Classification

 Q1:  What's purpose of part of Clustering? and how to implement it?

 Q2:  What does Object Classificaton based on?  based on target's width and height?  I think we can only get simple classification,e.g., large car, medium car and small car. Is it possible to distinguish human from car based on current TI technology?

     

  • 0
    TI__Expert 4720 points
    Hi Tianxi,

    Q1: What's purpose of part of Clustering? and how to implement it?

    A1: In general, scenes have a small number of large objects (for example, a highway scene would have say 10 cars). Higher layer processing is not interested in the 'point cloud' that detection algorithms would generate. It would rather have information that there are 10 objects in the scene moving with different velocities. The purpose of clustering is to take the detection output (which can be large - say 100 detected objects, after DoA) and reduce it to 10 relevant clusters. Once these clusters are created the size of the object would also become apparent. As far as implementation is concerned, there are different algorithms k-means, dBScan, etc, and you can implement them on the C674x DSP or the MSS.

    Q2: What does Object Classification based on? based on target's width and height? I think we can only get simple classification,e.g., large car, medium car and small car. Is it possible to distinguish human from car based on current TI technology?

    A2: Object classification can be based on different information collected from the target. They can include.
    1. Velocity
    2. The size of a cluster.
    3. The RCS of the target (Higher the RCS, the more likely that the target is metallic), etc.

    Regarding your question on distinguishing humans from cars, you can compare the RCS of a car, versus the RCS of a human, and set a threshold based on that. That is, if the RCS is smaller than a given threshold, it is a person, and if it greater it is a car. In an actual implementation, you'd have to take into account the cluster size, the velocity of the target etc, to make a determination. Note that TI doesn't currently offer an object Identification routine or a clustering routine.

    Regards,
    Anil