Spatially and semantically diverse points extraction using hierarchical clustering

Abstract

Diverse points extraction is an important process in the fields of location-based services and automated driving, among others. While existing research has investigated the selection of semantically diverse locations, the selection of points of interest that are both semantically and spatially diverse may be required in some scenarios. For instance, a diverse points path can be used to simulate automated driving (AD) on various types of roads. We propose a hierarchical clustering process involving graph spectral clustering and mixed data clustering, followed by a diversity selection algorithm. We evaluated our proposed method using OpenStreetMap road network data and determined that the method extracts points with improved semantic diversity compared with random selection. In addition, the proposed method also improved spatial diversity compared with when graph clustering, mixed data clustering and random selection were used. Thus, the proposed method provided an optimal balance for achieving the dual goals of semantic and spatial diversity.