A Novel Method to Select Hyperparameters of the DBSCAN Algorithm for RADAR Applications

As the automotive industry is coming up with various ADAS solutions, RADAR is playing an important role. There are many parameters concerning RADAR detections to acknowledge. Unsupervised Clustering methods are used for RADAR applications. DBSCAN clustering method which is widely used for RADAR applications. The existing clustering DBSCAN is not aligned very well with its hyperparameters such as epsilon (the radius within which each data point checks the density) and minimum points (minimum data points required within a circle to check for core point) for which a calibration is needed. In this paper, different methods to choose the hyperparameters of DBSCAN are compared and verified with different clustering evaluation criteria. A novel method to select hyperparameters of the DBSCAN algorithm is presented with the paper. For testing the given algorithm, ground truth data is collected, and the results are verified with MATLAB-Simulink.