Object Based Analytics for Finding Homogeneous Collections in Aerial Imagery Datasets

With the present technologies enabling automated detections in overhead imagery, detected objects can be further utilized for various analytical insights. The geological location information can enable spatial analysis of the objects. Multiple adjacent images in a region contribute detected objects over the cumulative spatial area. A homogeneous collections is defined as a group of objects from a single class that are closer than a distance, but also maintain a minimum separation. Examples are a group of aircrafts excluding junkyards, vessels at sea excluding those moored at shore, etc. Such insights are useful in applications such as urban planning and study of economic activity helping generate insights towards the Sustainable Development Goals (SDG). Two approaches to the implementation of proposed algorithm for finding homogeneous collections are presented and compared. Its efficacy is demonstrated on the xView 60-class dataset.