Automatic Road Detection - the Good and the Bad

OpenStreetMap - Charlottesville

Yesterday Steve Coast announced that Bing had released a new tool for doing automatic road detection using satellite imagery. The concept is definitely interesting as it provides a way to rapidly generate road data over the entire globe without need of manual tracing.

However, I remarked that it was particularly interesting that Steve was working on this. Several years ago, when OpenStreetMap was still an ambitious but unproven concept many people argued that road detection was a useful, and perhaps necessary, mechanism for actually capturing all the road data. Steve was quite adamant that while it was possible - and he demonstrated it - it wouldn't work for other reasons.

OpenStreetMap is more than just a set of lines that render to nice maps. It is a topologically connected, classified and attributed, labeled network of geographic entities. Each road consists of intersections, road classifications, names, speed limits, overpasses, and lanes. OpenStreetMap has provided a very rich set of linked, geographic data.

And beyond the data, it has built a community of invested members that careful capture, annotate, and cultivate the data in OpenStreetMap. This means that the data is captured, but also updated and maintained (ideally) with new information, changes, and other entities such as parks, buildings, bus stops and more.

So Steve convincingly pointed out that automatic road identification was interesting, it would circumvent these other benefits of what OpenStreetMap was working on: rich connected data, and a community of volunteers that would build and maintain the dataset. Road detection has a tendency to generate a large amount of data in an area that no one is actively working on the data. So you can gain what appears to be good coverage but limited local knowledge on intersections, names, and other metadata.

I don't think that these are insurmountable problems. The act of capturing GPS data can be tedious, inaccurate, or not readily possible in remote areas. Road detection can provide this data and users can work afterwards to improve the data, either remotely or using even simpler mobile devices that a user can annotate features without having to capture the entire geographic road line.

So my comment the other day was about pointing out an interesting change in message and strategy. I applaud the work of Steve and the Bing team in developing new tools, but there are many other pieces that warrant consideration. Steve even asked often if the bulk import of the TIGER/Line data was good or bad for the US community. In the end, I believe it was the right thing as it provided a canvas of data using open data that provided a validity to skeptics that OpenStreetMap was viable and valuable.

Now that OpenStreetMap has become increasingly adopted by the world's largest providers and users of data it is time to evaluate new tactics for gathering and maintaining data. However this can't be at the expense of what made OpenStreetMap a success for the past 5+ years.


About this article

written on
posted in OpenStreetMap Back to Top

About the Author

Andrew Turner is an advocate of open standards and open data. He is actively involved in many organizations developing and supporting open standards, including OpenStreetMap, Open Geospatial Consortium, Open Web Foundation, OSGeo, and the World Wide Web Consortium. He co-founded CrisisCommons, a community of volunteers that, in coordination with government agencies and disaster response groups, build technology tools to help people in need during and after a crisis such as an earthquake, tsunami, tornado, hurricane, flood, or wildfire.