Using iterated local alignment to aggregate trajectory data into a traffic flow map

TL;DR

The paper tackles the problem of noisy GNSS trajectories hindering high-resolution traffic-flow maps at the road level. It introduces an iterative local alignment framework that uses line blending to align nearby road segments to local references, producing flow maps on road segments rather than raster grids. Through synthetic and empirical experiments, the approach demonstrates improved spatial resolution and accuracy over traditional map-matching and raster-based methods, with quantified trade-offs for tuning parameters. The method is shown to be offline-feasible for moderate data sizes and is accompanied by publicly available data and interactive web maps, enabling multi-scale GIS-enabled traffic analysis.

Abstract

Vehicle trajectories are a promising GNSS (Global Navigation Satellite System) data source to compute multi-scale traffic flow maps ranging from the city/regional level to the road level. The main obstacle is that trajectory data are prone to measurement noise. While this is negligible for city level, large-scale flow aggregation, it poses substantial difficulties for road level, small-scale aggregation. To overcome these difficulties, we introduce innovative local alignment algorithms, where we infer road segments to serve as local reference segments, and proceed to align nearby road segments to them. We deploy these algorithms in an iterative workflow to compute locally aligned flow maps. By applying this workflow to synthetic and empirical trajectories, we verify that our locally aligned flow maps provide high levels of accuracy and spatial resolution of flow aggregation at multiple scales for static and interactive maps.

Figures (11)

• Figure 1: Trajectory data in Hannover, Germany. (a) Empirical trajectories (green circles), Loebensteinstraße (solid cyan circle). (b) Zoom of trajectories and map matched routes (blue lines). (c) Zoom of ideal flow aggregation. (d) Zoom of existing flow aggregation. (e) Zoom of 2 m $\times$ 2 m rasterised flow aggregation. Colour scale/label is traffic flow in road segments/grid cells.
• Figure 2: Unsplit/unsnapped and node split/node snapped routes. (a) Unsplit and unsnapped routes. (b) Node split and node snapped, with snap tolerance $\varepsilon_S=4$ m and with flow aggregation. Colour scale/label is traffic flow in road segments.
• Figure 3: Line blending leads to correct flow aggregation. (a) Before line blending and flow aggregation. Reference linestring $(7,ABC)$ in blue, candidate linestring $(2,AD)$ in orange. (b) After blending candidate into reference linestring, with blend tolerance $\varepsilon=4$ m, and flow aggregation. Blended linestring is $(9,AC'B)$.
• Figure 4: Snap candidate-touching linestrings to reference linestring after line blending. (a) Before line blending and snapping. Reference linestring $(7,AC)$ in blue, candidate linestring $(2, AC)$ orange, and candidate touching linestring $(5,CD)$ purple. (b) After line blending (blend tolerance $\varepsilon=4$ m) and snapping (snap tolerance $\varepsilon_S=4$ m). Reference linestring becomes $(9,AC'B)$, and candidate-touching linestring $(5,BD)$.
• Figure 5: Hannover routes line blending. (a) Flow segments without line blending. (b) Flow segments with line blending. Colour scale/label is traffic flow.