← Tutorials

Tutorial Series · Part 25 of 27

Map Matching and Feature Matching

Explain using roads, gravity, magnetic fields and landmarks as navigation information.

By

  • map-matching
  • feature-matching
  • aiding
Series Context

Navigation from First Principles

This article is part of an ordered course. Use the previous and next links to stay in sequence.

Progress Part 25 of 27
Previous Gravity and Navigation: Is Gravity Just 9.81? Next Building a Browser-Based Navigation Lab

Navigation systems can use the environment as a sensor. A car can match a noisy trajectory to a road network. A pedestrian system can use steps and floor plans. Advanced systems can use visual landmarks, terrain, gravity or magnetic signatures. Groves’ book treats environmental feature matching as part of the wider navigation-system landscape.1

Why road context helps

Suppose GNSS says a car is somewhere near two parallel roads. Which road is it on? A map-matching algorithm uses the position estimate, uncertainty, heading, speed and road geometry to infer the most likely path.

A simple likelihood view

For candidate road segment jj, define a residual distance:

rj=d(p^,roadj)r_j = d(\hat{\mathbf{p}}, road_j)

A simple likelihood can be:

Lj∝exp⁡(−rj22σ2)L_j \propto \exp\left(-\frac{r_j^2}{2\sigma^2}\right)

Add heading consistency:

Lj∝exp⁡(−rj22σp2)exp⁡(−Δψj22σψ2)L_j \propto \exp\left(-\frac{r_j^2}{2\sigma_p^2}\right) \exp\left(-\frac{\Delta\psi_j^2}{2\sigma_\psi^2}\right)

Then choose the most likely road or maintain multiple hypotheses.

Interactive demo: map-matching playground

The demo below projects noisy measurements onto nearby segments and ranks candidate roads by distance and heading agreement. It is still local and simplified, but it shows the ambiguity clearly when two roads are plausible.

Interactive Demo

Map-Matching Playground

Noisy GNSS samples are compared against nearby road candidates. The demo shows why map matching is a ranking problem, not a guarantee.

Grey roads are candidate segments. Blue dots are GNSS measurements.

Controls

Start with parallel roads and low noise. Then raise GNSS or heading noise until multiple road candidates become plausible.

Best candidates

1-
2-
3-

What to watch

  • With low noise, the best road dominates the candidate list.
  • Near parallel roads, the top two hypotheses can become genuinely close.
  • Turning multi-hypothesis mode off forces a single answer even when uncertainty is obvious.
  • Outliers can drag the measurement cloud away from the road, but continuity and probability still matter.

Try this

Use the parallel-road layout and increase GNSS noise until the dots overlap both roads. The right lesson is not that map matching is magic, but that it encodes a likelihood judgement using map context.

When map matching fails

Map matching fails if the map is wrong, incomplete or too restrictive. Feature matching can also produce false matches. Environmental information helps, but it has to be fused with uncertainty rather than treated as ground truth.

Footnotes

  1. Paul D. Groves, Principles of GNSS, Inertial, and Multisensor Integrated Navigation Systems, Second Edition, Artech House, 2013. https://uk.artechhouse.com/Principles-of-GNSS-Inertial-and-Multisensor-Integrated-Navigation-Systems-Second-Edition-P1557.aspx - Core textbook reference for navigation systems, GNSS, INS, multisensor integration, coordinate frames, dead reckoning, feature matching and environmental aiding. Accessed 2026-06-11. ↩