Terra.js - A semantic terrain engine

Felix Palmer | @pheeelicks | www.pheelicks.com

Terra.js - a semantic terrain engine

Felix Palmer | @pheeelicks | www.pheelicks.com

Talk overview

What is it?
Why build this?
Architecture overview
Plans for the future
Use cases

A semantic terrain engine

Why "Semantic"?
Rather than focussing on high resolution photography, understand landscape better
Generate a model from the data, rather than using them directly
Key is having land cover data in addition to satellite photography

DEMO

Warning: alpha quality, feel free to use your imagination

Why build this?

Originally motivated by visualizing train locations (AKA yak shaving)
Perhaps there is a use for a more general purpose framework?
Thought it might be fun

Towards a photo-realistic, realtime Earth

Live in the future, then build what's missing.

Paul Graham

1m accurate elevation and land cover data
Accurate weather data
Computers 100X faster than today with 100X internet bandwidth
Real-time stream of the earth's surface (see: Urthecast)
Immersive augmented reality (e.g. Oculus Rift, minus the nausea)

How do we get there?

Digitize everything
Understand everything

Quick recap

Terra.js is a framework for creating models representing places
The engine consumes data that semantically describes a place
A 3D model is generated by the engine

How does this work?

Data sources
Tile loading
Terrain generation on the GPU
Land cover + satellite imagery to create final image

Data

Basic building blocks are:
1. Aerial photography
2. Elevation data
3. Land cover data

Data sources

Satellite imagery
Elevation data
Land cover data

Data courtesy of:

NASA / ESA (GLOBCOVER)

Tile loading

Data sources are processed into tiles
Numerous zoom levels are produced
Based on the current camera location Terra.js downloads required tiles
Loaded tiles are passed to GPU for use

Terrain generation

Primitive geometry is produced in JavaScript and passed to the GPU
Vertex shader reads in elevation data and displaces vertices
Complication: Level of Detail

Rendering

Deformed terrain is passed to the fragment shader
Here GPU makes decisions on how to render based on land cover data
Satellite imagery is used to supplement generated model

Build what's missing

Better quality data, pixellation is obvious close to surface
Weather display
Rendering based on time of day
Seasonal rendering
Styling, cell shading etc
Procedural cities
Algorithmic botany

Algorithmic botany

Enables creation of detailed environments
Not necessary to stream all data, just land cover
After all, a tree's a tree

Image courtesy of algorithmicbotany.org

Algorithmic botany

Image courtesy of algorithmicbotany.org

Use cases

Train locations! (That yak won't shave itself)
National parks
Data visualization? Does 3D help here?
Games. Flight sims? Age of Empires anyone?
...?

Questions?

Felix Palmer | @pheeelicks | www.pheelicks.com