Processing Earth Observing images with Ames Stereo Pipeline

ICESat with its GLAS instrument provided valuable elevation measurements of glaciers. The loss of this spacecraft caused a demand for alternative elevation sources. In response to that, we have improved our Ames Stereo Pipeline (ASP) software (version 2.1+) to ingest satellite imagery from Earth satellite sources in addition to its support of planetary missions. This enables the open source community a free method to generate digital elevation models (DEM) from Digital Globe stereo imagery and alternatively other cameras using RPC camera models. Here we present details of the software. ASP is a collection of utilities written in C++ and Python that implement stereogrammetry. It contains utilities to manipulate DEMs, project imagery, create KML image quad-trees, and perform simplistic 3D rendering. However its primary application is the creation of DEMs. This is achieved by matching every pixel between the images of a stereo observation via a hierarchical coarse-to-fine template matching method. Matched pixels between images represent a single feature that is triangulated using each image's camera model. The collection of triangulated features represents a point cloud that is then grid resampled to create a DEM. In order for ASP to match pixels/features between images, it requires a search range defined in pixel units. Total processing time is proportional to the area of the first image being matched multiplied by the area of the search range. An incorrect search range for ASP causes repeated false positive matches at each level of the image pyramid and causes excessive processing times with no valid DEM output. Therefore our system contains automatic methods for deducing what the correct search range should be. In addition, we provide options for reducing the overall search range by applying affine epipolar rectification, homography transform, or by map projecting against a prior existing low resolution DEM. Depending on the size of the images, parallax, and image quality, one of these methods will perform better than the others. Because our software requires minimal user input in the form of command line arguments and has no graphical user interface (GUI), it is uniquely adept for clustering computer environments as well as normal workstations. Users can run multiple stereo sessions simultaneously to process a large number of stereo pairs. We also provide a tool called ';parallel_stereo' which utilizes GNU's parallel command to split a single stereo pair session across multiple nodes that have a shared file system. This greatly reduces the processing time of a single stereo pair. Our primary system of testing was NASA's Pleiades super computer that is freely available to NASA sponsored scientists and engineers via the High End Computing Capability (HECC) project. ASP is freely available for download from our website at irg.arc.nasa.gov/ngt/stereo. Source code for all releases and current development code is available from our Github account at github.com/NeoGeographyToolkit/StereoPipeline. In addition to being free, ASP is Apache 2 licensed which offers free license to copy, redistribute, and sell the software and results pending the license of the input imagery. It is our goal that this software will make DEMs and the processing of satellite imagery more accessible to scientists of all fields.