Terrain From Imagery

In 2009 the need for accurate high resolution terrain data is only growing. One can see increasing numbers of applications that use terrain all over the geospatial industry and beyond. Part of the reason for the proliferation of terrain and applications that use it stems from increased usage in all methods of collecting and creating terrain. The SRTM mission provided course-resolution terrain for large portions of the world. LIDAR sensors continue to sell briskly, and of course automatically-generated terrain from triangulated oriented images (stereo pairs) continues to grow as increasing numbers of sensors capture image data.

Automatic terrain generation, which involves producing TINs and grids by correlating XYZ points from a pair of overlapping oriented images, is nothing new. In the context of softcopy photogrammetry the technology is over a decade old, and at ERDAS we have produced automatic terrain extraction solutions since the release of OrthoBase Pro back in 2001. Since then most photogrammetry practitioners have worked automatic terrain generation into their production workflows, particularly as a source for orthophoto generation. While I believe sensor fusion (capturing optical and LIDAR data simultaneously) will have an increasing role in the future, the reality is that automatically-generated terrain is also likely to play an important role for years to come. Why? Because data is becoming captured at increasingly high resolution and software solutions are evolving in their ability to generate corresponding high-density terrain data. A good example is the ADS80 sensor from Leica Geosystems. It can capture imagery at a 5 centimeter pixel resolution. Data captured at this resolution allows for very dense (and accurate) automatically extracted terrain.

In the photogrammetry group at ERDAS we’ve been putting effort into new techniques for automatic terrain extraction from imagery. Specifically we are looking at automating as much as possible, updating terrain extraction algorithms, and supporting modern IT infrastructures by adding distributed processing capability. After all, generating massive terrain datasets can require some serious computer power.

The image below shows terrain with a 10 centimeter density (post spacing) processed as an LAS file and displayed in GeoCue's PointVue LE viewing software (designed for visualizing LIDAR data). Color is based on elevation, so you can easily see the buildings and other surface features. Using a LIDAR viewer as a QC tool helps because of the density of the data.

In the image below I've rotated the view and applied intensity shading. This displays the detail quite well: it is possible easily discern the buildings, roads, crops and vegetation.

And a view of another area: