Communication-efficient learning of traffic flow in a network of wireless presence sensors
Current traffic management systems learn global traffic flow models based on measurements from a static mesh of hard-wired presence sensors. The centralization of all data comes at the cost of limited scalability, security and fault-tolerance. Modern traffic control could benefit from a decentralized system of cheap wireless sensors. However, con-strained devices pose challenges for data analysis, which must be communication- and energy-efficient as well as secure. We hereby present a privacy-preserving decentralized in-network algorithm which exchanges space-time aggregated values between restricted sets of topological neigh-boring nodes. The algorithm’s evaluation on real world traffic data demonstrates its performance in terms of communication cost and accuracy.
READ FULL TEXT