CODEWITHSUNDEEP
matlabdistanceeuclidean-distancedbscan
Metz
Metz

Reputation: 73

Ellipse distance metric for DBSCAN clustering

I am using the DBSCAN algorithm to determine clusters in a data set obtained by an automotive radar. The paper "Grid-Based DBSCAN for Clustering Extended Objects in Radar Data" from Dominik Kellner, Jens Klappstein and Klaus Dietmayer (link below) proposes a Grid-Based DBSCAN method. Therefore, the search radius epsilon variates in azimuth direction depending on the range. The radius in range direction stays constant. The normal DBSCAN is using Euclidean distance metric to determine the epsilon-neighbourhood where the search radius is the same in both directions. I cannot find out how to have an ellipse-search instead of a circular.

Do you know a distance metric that is working elliptical? Or, can you provide me with a short code that solves my problem? I am using MATLAB but the code can be in your prefered language.

Let's give an example so we talk about the same:

Consider a cartesian coordinate system with range in meters plotted against azimuth angle in degrees. The search distance in the range direction should be three meters (or possible observation points) in both directions from a centre point. In azimuth direction, the search radius should be five points in both directions.

If you cannot think of an elliptical solution, maybe a linear works as well.

Thank you for your help.

https://www.researchgate.net/profile/Dominik_Kellner2/publication/261127945_Grid-based_DBSCAN_for_clustering_extended_objects_in_radar_data/links/57742a7708aead7ba06e60b5.pdf

Upvotes: 1

Views: 689

Answers (0)

Related Questions