Map stream: Initializing what-if analyses for real-time symbiotic traffic simulations

Sunderrajan, Abhinav; Aydt, Heiko; Cai, Wentong; Knoll, Alois · 2014 · Crossref

DOI: 10.1109/wsc.2014.7020130

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 initializing city-scale symbiotic traffic simulations, which require real-time data from thousands of vehicles to perform what-if analyses. A critical bottleneck in this process is map-matching: aligning noisy, continuous GPS data streams with a digital road network to determine vehicle locations. Traditional batch-processing map-matching techniques are unsuitable for high-volume, low-latency streaming requirements. The authors propose a scalable, real-time map-matching algorithm designed to minimize latency while maintaining high accuracy, enabling the rapid initialization of simulation states. The proposed method utilizes a Hidden Markov Model (HMM) adapted for streaming data via sliding windows of size two. Instead of processing entire trajectories, the algorithm computes the most probable route based on two successive GPS samples. It calculates emission probabilities using Gaussian noise models and transition probabilities based on the discrepancy between the great-circle distance of GPS points and the shortest path length on the road network. To reduce computational latency, the authors implement two key optimizations: partitioning the road network using QuadTrees to limit the search space for candidate edges, and employing an arc-flag approach to accelerate shortest-path computations during transition probability estimation. Experiments were conducted using synthetic GPS data generated from the Singapore road network, testing various sampling intervals (1 to 70 seconds) and Gaussian noise levels (0 to 14 meters standard deviation). Results indicate that reliability, defined as the percentage of correctly mapped road segments, ranges from 85% to 92% when sampling intervals are between 10 and 40 seconds and noise is between 5 and 9 meters. Latency analysis reveals that computation time is invariant to noise levels but increases with longer sampling intervals due to more complex shortest-path calculations. The study finds that sampling intervals of 10 to 40 seconds offer the optimal balance of accuracy and latency. Sampling intervals below 10 seconds result in excessive latency (exceeding 2 seconds), which would cause memory bottlenecks in high-velocity streams involving 100,000 vehicles. The significance of this work lies in providing a viable method for initializing real-time symbiotic simulations with minimal delay. By demonstrating that moderate sampling intervals yield high reliability with manageable latency, the paper supports the feasibility of large-scale, real-time traffic management systems. The authors conclude that a parallel implementation of this algorithm, potentially using stream-processing engines like STORM, is necessary to handle the volume of data from thousands of vehicles simultaneously. Future work includes refining transition probabilities using historical traffic data at road splits to further enhance accuracy.

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.

• in vehicle coaching