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.

Choice of the dbscan algorithm

Other Parts Discussed in Thread: AWR1843

Hello,

I'm using AWR1843 radar and I want to implement a clustering algorithm in the OOB demo.
In the other demos I see that the Dbscan algorithm is applied. I want to know why DBscan is chosen and not another clustering algorithm.

If there is another more efficient one, please share it with me and explain the reason of this selection?

Thank you,

Ben Hassine

  • Hello Ben,

    We found DBSCAN to be a good algorithm for finding clusters in radar point cloud and hence chose to use the same in our demos. It finds clusters of varying shapes (depending on the nature of point cloud spread which is subject to change with each new observation) based on the density of points across different subsections of the point cloud spread (two main parameters that define the density are Epsilon and the number of points). It performs really well on a noisy radar point cloud by clustering points within higher density regions and ignores the sparser sections which resemble noise.

    Regards,

    Kaushik 

  • Hello,

    Thank you for your reponse. And what about the K_means algorithm? According to my research, it is very fast compared to dbscan.
    it is true that we have to define the paramaetr K but there is the "Elbow method" to select the optimal number of clusters. I want to know if it is more efficient than dbscan and if we can implement it in the OOB demo.

    Regards,

    Ben Hassine

  • Hi Ben,

    DBSCAN can address the issue of irregular shaped clusters much better than K-means. Furthermore, K-means clustering will classify all points (including noise points and other points of no interest) on your dataset into their respective clusters.

    For more clarity I want to ask you the following:

    1. What was your use case? i.e., what type of dataset did you conduct your research on?
    2. Can you clarify what you mean by "more efficient"? Are you talking about an improvement in computation speed or better results with respect to how the clusters are classified?

    Regards,

    Kaushik

  • Hi Kaushik,

    Thank you very much for your reponse. 

    1) For the question 1  : I'm trying to cluster and classify "static" objects of "concrete" or "metallic" nature

    2) For the question 2 :  Yes, I mean by more efficient for two reasons :  an improvement in computation speed and better results with respect to how the clusters are classified.

    So is DBSCAN the best algorithm for my case or is there another better algorithm?

    Regards,

    Ben Hassine

  • Hi Ben,

    DBSCAN is certainly one of the better options for your use case, but I cannot make a statement that it is the best there is.

    Regards,

    Kaushik

  • Hi Kaushik,

    Thank you for your response that helped me a lot.

    Regards,

    Ben Hassine