Towards fine-grained urban traffic knowledge extraction using mobile sensing

Ban, Xuegang (Jeff); Gruteser, Marco · 2012 · Crossref

DOI: 10.1145/2346496.2346514

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 extracting fine-grained urban traffic knowledge from mobile sensing data, specifically GPS location traces. While previous research focused on city-scale metrics like travel times and routing, this work targets micro-level details such as intersection performance and signal timing. The motivation stems from the limitations of traditional fixed-location sensors (e.g., loop detectors), which are expensive to deploy and sparse in coverage, and the potential of mobile sensors to provide continuous, high-resolution data. However, utilizing mobile data requires overcoming two primary hurdles: the distinct format of mobile trajectories compared to fixed-sensor data, and significant privacy concerns regarding the collection of personal location information. To address these challenges, the authors propose a co-design framework that integrates privacy protection with novel traffic modeling techniques. For privacy, they employ a "privacy-by-design" approach using Virtual Trip Lines (VTLs). VTLs define specific zones along roadways where data collection is permitted, restricting data gathering to areas necessary for traffic modeling while omitting identifiers and limiting trace continuity to reduce re-identification risks. For modeling, the authors integrate traffic flow theory with learning and optimization algorithms. This hybrid approach leverages the physical principles of traffic behavior (such as shockwave theory) to interpret the unique features of mobile data, such as discontinuities in trajectories that indicate signal changes or queue formation. The paper illustrates this framework through two case studies. First, for estimating intersection delay patterns, the authors use traffic shockwave theory to model delay as piecewise linear curves. They apply least-square estimation to fit sample travel times from mobile sensors to these theoretical curves, allowing for the reconstruction of delay patterns and the identification of traffic conditions like over-saturation. Second, for estimating signal timing parameters (cycle length, red/green times), the authors develop a method using Support Vector Machines (SVM). This technique identifies "cycle-breaking vehicles" by analyzing arrival time differences and delay differences between consecutive sampled vehicles. The SVM model effectively separates cycle-starting vehicles from others, enabling accurate detection of cycle boundaries and effective red/green durations. Results from simulation and NGSIM datasets demonstrate significant improvements in accuracy, reducing root mean square error in cycle detection from 9.6 seconds to 1.6 seconds in simulations. The significance of this work lies in demonstrating that mobile sensing can replace or supplement fixed sensors for detailed traffic operations, provided that privacy and modeling are jointly optimized. The authors conclude that future research must expand these deterministic models to account for stochastic traffic nature and develop application-specific privacy frameworks. The proposed paradigm shifts traffic modeling toward a privacy-aware approach, enabling the extraction of rich, fine-grained traffic knowledge from the growing volume of mobile data while protecting user privacy.

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.

• exposure measurement