A Filtering-Based Approach for Improving Crowdsourced GNSS Traces in a Data Update Context

Abstract

Traces collected by citizens using GNSS (Global Navigation Satellite System) devices during sports activities such as running, hiking or biking are now widely available through different sport-oriented collaborative websites. The traces are collected by citizens for their own purposes and frequently shared with the sports community on the internet. Our research assumption is that crowdsourced GNSS traces may be a valuable source of information to detect updates in authoritative datasets. Despite their availability, the traces present some issues such as poor metadata, attribute incompleteness and heterogeneous positional accuracy. Moreover, certain parts of the traces (GNSS points composing the traces) are results of the displacements made out of the existing paths. In our context (i.e., update authoritative data) these off path GNSS points are considered as noise and should be filtered. Two types of noise are examined in this research: Points representing secondary activities (e.g., having a lunch break) and points representing errors during the acquisition. The first ones we named secondary human behaviour (SHB), whereas we named the second ones outliers. The goal of this paper is to improve the smoothness of traces by detecting and filtering both SHB and outliers. Two methods are proposed. The first one allows for the detection secondary human behaviour by analysing only traces geometry. The second one is a rule-based machine learning method that detects outliers by taking into account the intrinsic characteristics of points composing the traces, as well as the environmental conditions during traces acquisition. The proposed approaches are tested on crowdsourced GNSS traces collected in mountain areas during sports activities.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 11 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.