Automating Road Mapping From Aerial Images

Advances in mapping have perhaps not been one of the more heralded technological advances in recent years, but they have fundamentally changed the world.  Despite the tremendous progress that has been made however, there are still large chunks of the 20 million or so miles of roads around the world that have yet to be mapped.  When you think of the labor intensity of such work, it’s understandable.

A recent paper by researchers from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) documents a new approach to the job.  It describes a new system, called RoadTracer, which automatically generates highly accurate road maps using data from aerial images.

The team believe that their method is not only more accurate than existing approaches, but also significantly more cost-effective.  Both of these together open up parts of the world that might previously have been uneconomic to map.

“RoadTracer is well-suited to map areas of the world where maps are frequently out of date, which includes both places with lower population and areas where there’s frequent construction,” the team say. “For example, existing maps for remote areas like rural Thailand are missing many roads. RoadTracer could help make them more accurate.”

Automate mapping

There have been a number of attempts to automate mapping, many of which involve neural networks trained on aerial images labelled to identify roads and other features.  Such systems often require post-processing however to overcome incomplete and ambiguous aerial images.

These methods are far from perfect however as even slight mislabelling can be amplified in the final road map.  Such errors are also especially likely as aerial photos capture a complex landscape, especially when shadows or poor lighting obscure the picture.

By contrast, RoadTracer creates maps step-by-step, beginning at a known location in the road network.  It then utilizes a neural network to examine the area surrounding the road to determine the likeliest next point in the road.  Once that point has been added, it moves on until the entire road network is mapped out.

“Rather than making thousands of different decisions at once about whether various pixels represent parts of a road, RoadTracer focuses on the simpler problem of figuring out which direction to follow when starting from a particular spot that we know is a road,” the team say. “This is in many ways actually a lot closer to how we as humans construct mental models of the world around us.”

Training the system

The system was trained on aerial images from some 25 cities across six different countries in both North America and Europe.  Once trained, it was then tested on a further 15 cities.

“It’s important for a mapping system to be able to perform well on cities it hasn’t trained on, because regions where automatic mapping holds the most promise are ones where existing maps are non-existent or inaccurate,” the team say.

When the results of this test run were analyzed, the system had performed admirably, beating current systems by around 45%.  Whilst there is still clearly room for improvement, the results do suggest that this automated approach could provide clear advantages over manual systems, although even with RoadTracer, there is still an important role for humans in double-checking the output.

“That said, what’s clear is that with a system like ours you could dramatically decrease the amount of tedious work that humans would have to do,” the team conclude.

Suffice to say, aerial images are just one bit of the equation as they won’t provide researchers with information on roads, such as overpasses and underpasses, that are impossible to ascertain from above.  To overcome this problem, the team are working on algorithms that can create maps from GPS data, with the eventual goal of merging the two systems together into a complete mapping application.

It’s a fascinating project, and you can learn more about it via the video below.