In-Network Approximate and Efficient Spatiotemporal Range Queries on Moving Objects

Abstract

Data aggregations enable privacy-aware data analytics for moving objects. A spatiotemporal range count query is a fundamental query that aggregates the count of objects in a given spatial region and a time interval. Existing works are designed for centralized systems, which lead to issues with extensive communication and the potential for data leaks. Current in-network systems suffer from the distinct count problem (counting the same objects multiple times) and the dead space problem (excessive intra-communication from ill-suited spatial subdivisions). We propose a novel framework based on a planar graph representation for efficient privacy-aware in-network aggregate queries. Unlike conventional spatial decomposition methods, our framework uses sensor placement techniques to select sensors to reduce dead space. A submodular maximization-based method is introduced when the query distribution is known and a host of sampling methods are used when the query distribution is unknown or dynamic. We avoid double counting by tracking movements along the graph edges using discrete differential forms. We support queries with arbitrary temporal intervals with a constant-sized regression model that accelerates the query performance and reduces the storage size. We evaluate our method on real-world mobility data, which yields us a relative error of at most 13 . 8% with 25 . 6% of sensors while achieving a speedup of 3 . 5 × , 69 . 81% reduction in sensors accessed, and a storage reduction of 99 . 96% compared to finding the exact count.