STREAM: System for Trajectory Reference Encoding And Modeling.
Authors
Vilslev, Daniel ; Lykkegaard, Martin Opal
Term
4. term
Education
Publication year
2023
Submitted on
2023-06-15
Pages
18
Abstract
This thesis tackles the challenge of compressing ship GPS data in unrestricted environments, where movement is not confined to road networks. It introduces STREAM, a system that combines model-based compression with a novel reference model. The reference model builds a reference set from historical trajectories by splitting them into sub-trajectories, indexing them with an R-tree, and selecting representative references based on overlap. During compression of new data, the model incrementally matches against a reference within a specified error bound and stores only the reference id with start and end indices, or a count when many points can be represented by the same reference point. STREAM’s architecture comprises an edge-based compressor and a cloud component that would store data, compute new reference trajectories, and perform decompression; in this work, the cloud remains conceptual since the compressor suffices for proof of concept. STREAM employs both a vector-based model from prior work and the proposed reference model, choosing between them to optimize compression or runtime. Experiments indicate that STREAM achieves more than twice the compression ratio of the baseline MOBY and offers greater flexibility, suggesting clear improvements for maritime GPS compression.
Denne afhandling adresserer udfordringen med at komprimere GPS-data fra skibe i ubegrænsede netværk, hvor bevægelser ikke er bundet til vejnet som ved biler. Den foreslår STREAM, et system der kombinerer modelbaseret kompression med en ny referencebaseret model. Referencemodellen konstruerer et referencesæt fra historiske data ved at opdele ruter i sub-trajektorier, indeksere dem i en R-tree-struktur og udvælge repræsentative referencer baseret på overlap. Ved komprimering af nye data matcher systemet løbende mod en reference inden for en given fejlgrænse og gemmer kun reference-id samt start- og slutindeks eller en tælling, når mange punkter kan repræsenteres af samme referencepunkt. STREAMs arkitektur omfatter en edge-baseret kompressor og en cloud-del, hvor cloud skal lagre data, beregne nye referenceforløb og stå for dekomprimering; i denne afhandling er cloud-delen holdt på et teoretisk niveau, da beviset for konceptet kan evalueres på kompressoren alene. STREAM bruger både en vektorbaseret model fra tidligere arbejde og den foreslåede referencemodel og kan vælge imellem dem for at optimere kompression eller køretid. Eksperimenter viser, at STREAM opnår mere end dobbelt så høj kompressionsratio som baseline-systemet MOBY og leverer større fleksibilitet, hvilket peger på klare forbedringer for maritim GPS-komprimering.
[This apstract has been generated with the help of AI directly from the project full text]
Keywords