Centralized Algorithms Based on Clustering with Self-tuning of Parameters for Cooperative Target Observation

Abstract

Clustering on target positions is a class of centralized algorithms used to calculate the surveillance robots' displacements in the Cooperative Target Observation (CTO) problem. This work proposes and evaluates Fuzzy C-means (FCM) and Density-Based Spatial Clustering of Applications with Noise (DBSCAN) with K-means (DBSk) based self-tuning clustering centralized algorithms for the CTO problem and compares its performances with that of K-means. Two random motion patterns are adopted for the targets: in free space or on a grid. As a contribution, the work allows identifying ranges of problem configuration parameters in which each algorithm shows the highest average performance. As a first conclusion, in the challenging situation in which the relative speed of the targets is high, and the relative sensor range of the surveillance is low, for which the existing algorithms present a substantial drop in performance, the FCM algorithm proposed outperforms the others. Finally, the DBSk algorithm adapts very well in low execution frequency, showing promising results in this challenging situation.