Heterogeneous urban traffic data and their integration through kernel-based interpolation

Chow, Andy · 2016 · Crossref

DOI: 10.1108/jfm-08-2015-0025

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the challenge of integrating heterogeneous urban traffic data to reconstruct detailed spatio-temporal traffic patterns. While previous research has focused on freeway data using model-based approaches like Kalman Filters, urban networks present greater complexity due to diverse data sources with varying granularity, latency, and accuracy. The authors propose a kernel-based interpolation method that fuses data from different sensors—such as fixed loop detectors, Global Positioning System (GPS) devices, and Automatic Number Plate Recognition (ANPR) cameras—without requiring an underlying traffic flow model. This approach aims to provide a cost-effective tool for offline transport planning and policy evaluation by overcoming the difficulties associated with selecting and calibrating complex traffic models. The methodology consists of two primary steps: data smoothing and data integration. First, the algorithm smooths raw data onto a common space-time grid using kernel-based interpolation. To account for the anisotropic nature of traffic flow, where characteristics propagate differently in free-flow versus congested conditions, the method employs distinct kernel functions for each state. These are combined using a weighting factor based on a hyperbolic tangent function, which transitions between free-flow and congested estimates based on a speed threshold. Sectional journey time data, such as that from ANPR cameras, is converted into point measurements by constructing virtual vehicle trajectories. Second, the smoothed data fields from various sources are integrated using a weighted linear combination, where weights reflect the accuracy and reliability of each source. The performance of the algorithm was evaluated using a virtual test-bed created with VISSIM microscopic simulation software. The simulation replicated a section of Waterloo Road in Central London, generating a high-resolution "ground truth" speed field. Synthetic data mimicking GPS probes (with a 6.6% penetration rate) and ANPR journey times were derived from the simulation to test the fusion algorithm's ability to reconstruct the traffic state. Additionally, the paper demonstrates the method’s applicability through a real-world case study using actual traffic data collected from the same road section in London, highlighting the differences in spatial and temporal resolution between ANPR and GPS data sources. The study concludes that the kernel-based interpolation approach effectively integrates heterogeneous data sources to produce fine-grained traffic patterns. By avoiding the need for explicit traffic flow models, the method offers a robust alternative to traditional filtering techniques, particularly for offline applications. The results demonstrate that combining sparse but spatially detailed GPS data with more frequent but spatially coarse ANPR data yields a comprehensive view of urban traffic dynamics. This contributes to the field of facilities management and transport network analysis by providing a reliable means to utilize big data for understanding congestion and evaluating transport policies.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• telematics crash prediction