Trajectory Map-Matching in Urban Road Networks Based on RSS Measurements

TL;DR

This work tackles the problem of reconstructing vehicle trajectories in urban road networks using only RSS measurements, without GPS. It introduces a Hidden Markov Model-based RSS embedding (HRE) that alternates between learning the signal-propagation model parameters and inferring the trajectory, augmented by a fast maximum-speed-based rough trajectory estimator (MSR) for reliable initialization. An adaptive mobility extension (HREA) and a two-stage graph-based trajectory reconstruction further enhance robustness to speed variability and data sparsity. Experiments on two real-city datasets (Chengdu and Shenzhen) show that HRE/HREA achieve sub-10 m queueing-error and sub-0.5% trajectory-matching error, outperforming traditional map-matching baselines and maintaining performance under up to 30% data missing, illustrating practical GPS-free trajectory inference in dense urban wireless environments.

Abstract

This paper proposes an RSS-based approach to reconstruct vehicle trajectories within a road network, enforcing signal propagation rules and vehicle mobility constraints to mitigate the impact of RSS noise and sparsity. The key challenge lies in leveraging latent spatiotemporal correlations within RSS data while navigating complex road networks. To address this, we develop a Hidden Markov Model (HMM)-based RSS embedding (HRE) technique that employs alternating optimization to infer vehicle trajectories from RSS measurements. This model captures spatiotemporal dependencies while a road graph ensures network compliance. Additionally, we introduce a maximum speed-constrained rough trajectory estimation (MSR) method to guide the optimization process, enabling rapid convergence to a favorable local solution.

Figures (8)

• Figure 1: Illustration of the relationship between location space and space: The relationship indicated by yellow arrows is modeled by propagation, while the relationship depicted by green arrows is modeled by vehicle mobility.
• Figure 2: The frequency histogram of real speeds and recovered speeds v.s. the of the speed model.
• Figure 3: (a) Relationship between trajectory length and the runtime of Algorithm \ref{['alg:overal']} on Dataset I. The figure displays the mean values (depicted by points) from 200 runs of randomly generated trajectories with length $T$, accompanied by $\pm 3\sigma$ confidence intervals. (b) Convergence behavior of Algorithm \ref{['alg:overal']} on Dataset I and II.
• Figure 4: Relationship between TME performance and missing rate on Dataset I. The figure displays the mean values (depicted by points) from 200 runs with data missing at each missing rate, accompanied by $\pm 3\sigma$ confidence intervals.
• Figure 5: Example trajectory estimation result of our HRE with different missing rates $r_{\text{M}}$. The red pentagons represent , while the black dots indicate the estimated vehicle positions using our HRE method.