Match Scribbles to Features on the Earth


Land Lines is an experimental online tool that matches drawn gestures on the screen with landforms extracted from satellite imagery. Developed by  Zach Lieberman, Matt Felsen, and the Data Arts Team at Google, Land Lines lets users explore Google Earth satellite imagery through the drawing of lines.

The tools works only in Chrome and provides imagery matches to squiggles whether the user draws it on a desktop or a mobile device by harnessing the combination of machine learning, optimized algorithms, and graphics card power.

In developing the tool, OpenCV Structured Forests and ImageJ’s Ridge Detection were used to determine dominant lines from an initial subset of over 50,000 satellite imagery tiles. Those lines were then stored in a vantage point tree which allows the tool to rapidly return an imagery match to an existing land feature in response to a user’s doodle.  To get into more technical detail about Land Lines, read the technical case study or browse the open-source code on GitHub.

The length of the line drawn is limited to a small length.  As you continue to draw, the line thinks from the starting point, keeping a consistent length.  As with a lot of experimentation, Land Lines can return remarkably accurate matches:

and some not so accurate matches:

If you just want to see where the line takes you, switch to the Drag function.  From there you can be taken on an endless journey of connected lines.  As each tile is placed, a label letting you know the geographic location is provided.