[Paper Review] Map matching when the map is wrong: Efficient vehicle tracking on- and off-road for map learning

Jamie Murphy, Yuanyuan Pao+1 more|arXiv (Cornell University)|Sep 25, 2018

Automated Road and Building Extraction26 references4 citations

TL;DR

This paper proposes a semi-interacting multiple model (sIMM) filter that unifies on-road map matching with off-road free-space tracking to improve trajectory accuracy when maps contain errors or omissions. By allowing on-road HMM-based tracking to fall back to a closed-form off-road filter when map constraints are violated, the method enables robust, real-time vehicle tracking and detects map errors such as missing roads, incorrect turn restrictions, and misdirected one-way streets.

ABSTRACT

Given a sequence of possibly sparse and noisy GPS traces and a map of the road network, map matching algorithms can infer the most accurate trajectory on the road network. However, if the road network is wrong (for example due to missing or incorrectly mapped roads, missing parking lots, misdirected turn restrictions or misdirected one-way streets) standard map matching algorithms fail to reconstruct the correct trajectory. In this paper, an algorithm to tracking vehicles able to move both on and off the known road network is formulated. It efficiently unifies existing hidden Markov model (HMM) approaches for map matching and standard free-space tracking methods (e.g. Kalman smoothing) in a principled way. The algorithm is a form of interacting multiple model (IMM) filter subject to an additional assumption on the type of model interaction permitted, termed here as semi-interacting multiple model (sIMM) filter. A forward filter (suitable for real-time tracking) and backward MAP sampling step (suitable for MAP trajectory inference and map matching) are described. The framework set out here is agnostic to the specific tracking models used, and makes clear how to replace these components with others of a similar type. In addition to avoiding generating misleading map matching trajectories, this algorithm can be applied to learn map features by detecting unmapped or incorrectly mapped roads and parking lots, incorrectly mapped turn restrictions and road directions.

Motivation & Objective

• Address the critical limitation of standard map matching algorithms, which fail or produce incorrect trajectories when road maps contain errors or omissions.
• Enable accurate vehicle tracking in scenarios where GPS traces deviate from known road networks due to map inaccuracies.
• Develop a scalable, real-time tracking system that can detect and learn from map errors using large-scale GPS data.
• Provide a principled framework that unifies on-road HMM-based map matching with off-road free-space tracking for improved robustness.
• Enable map learning by identifying consistent off-road trajectory patterns that indicate missing or incorrect map features.

Proposed method

• Formulates a semi-interacting multiple model (sIMM) filter that combines on-road hidden Markov model (HMM) tracking with a closed-form off-road tracker, allowing fallback to off-road motion when on-road constraints are violated.
• Uses a forward filter for real-time tracking, enabling efficient online estimation of vehicle trajectories that switch between on-road and off-road states.
• Employs a backward MAP sampling step to compute maximum a posteriori (MAP) trajectories, improving accuracy by incorporating future observations.
• Introduces a model interaction constraint that avoids full IMM complexity by allowing the off-road filter to run independently while enabling on-road tracking to fall back to it.
• Designs the framework to be modular, allowing interchangeable on-road HMM trackers and off-road free-space filters without altering core inference logic.
• Leverages trajectory consistency and geometric constraints to detect map errors such as missing roads, incorrect turn restrictions, and misdirected one-way streets.

Experimental results

Research questions

• RQ1 How can map matching be made robust to incomplete or incorrect road network maps without degrading tracking accuracy?
• RQ2 Can a unified framework efficiently combine on-road HMM-based map matching with off-road free-space tracking in real time?
• RQ3 To what extent can off-road trajectory segments detected by the system indicate map errors such as missing roads or incorrect turn restrictions?
• RQ4 How does the performance of the sIMM filter compare to standard HMM-based map matching in terms of runtime and accuracy under map errors?
• RQ5 Can large-scale GPS traces be used to automatically detect and suggest corrections for map errors using the proposed system?

Key findings

• The sIMM filter achieves robust trajectory estimation even when maps contain missing roads, incorrect turn restrictions, or misdirected one-way streets, avoiding the erroneous on-road trajectories produced by standard map matching.
• The method detects map errors with high confidence by identifying consistent off-road trajectory patterns across multiple GPS traces, such as repeated off-road motion in areas where roads are missing.
• In the Presidio of San Francisco, the system detected a large number of off-road traces that corresponded to a newly constructed road not present in the OSM map as of January 2018, later confirmed by satellite imagery.
• The runtime of the sIMM trajectory generation is two to four times slower than standard HMM-based map matching, but this overhead is acceptable for web-scale applications and is expected to improve with further optimization.
• The framework enables map learning by identifying recurring off-road motion patterns that signal map inaccuracies, such as missing parking lots or incorrectly mapped road directions.
• The system successfully reconstructs trajectories that are more accurate than standard map matching when maps are incomplete or incorrect, particularly in areas with unmapped or misclassified infrastructure.