Review of SOCET GXP v4.0

A 1.791Mb PDF of this article as it appeared in the magazine complete with images is available by clicking HERE

Airborne LiDAR is used in municipal planning, the utility industry, for forest assessment, coastal mapping…the list could go on. The short of it is that LiDAR is clearly no longer the niche technology it once was, yet LiDAR data in its purest form–the point cloud, has been inaccessible to many geospatial professionals simply because the major software vendors offered, at best, limited support.

2012 is the year that has changed. Esri, Intergraph, and BAE Systems are all adding LiDAR support to their flagship products–ArcGIS, ERDAS IMAGINE and SOCET GXP, respectively. In this article I will present my review of SOCET GXP v4.0, the latest release from BAE Systems and the first version to add native support for LiDAR data.

SOCET GXP may not be as well known as some other geospatial software packages, but it has a rich history and is the de facto geospatial platform for many in the defense industry. SOCET GXP traces its lineage back to SOCET SET, a photogrammetric software package that for many years served as an industry standard for orthophoto production. SOCET SET was powerful, but a bit cumbersome to use in the fast-paced environment of imagery intelligence, and thus SOCET GXP was born to meet the needs of defense imagery analysts. Over the years BAE Systems has added a suite of capabilities to extend SOCET GXP from a tool for defense imagery analysts to a robust geospatial software package with photogrammetric, image processing, full-motion video, and feature collection capabilities, among others.

Within SOCET GXP data can be viewed within the standard 2D viewer, called the Multiport, or the new 3D Multiport. LiDAR point clouds can be loaded into the 3D Multiport where they can be fused with imagery and vector data, or converted to terrain files or intensity images for use in the 2D Multiport.

My first test was to load several LAS files into the 3D Multiport. Users should note that SOCET GXP does need to generate a temporary internal set of files for each LAS file loaded. These files are generally quite small, but for large data sets users should ensure that enough disk space exists. I found that point clouds containing millions of points loaded quickly. The controls for panning and zooming are modeled on Google Earth, which bodes well if you are a Google Earth user. While I find Google Earth’s controls just fine in a virtual globe environment, or when massive image data sets are loaded into the 3D Multiport, I felt the need for more precision than the controls offered.

SOCET GXP allows point clouds to be symbolized according to their elevation, return number, and classification. They can also be colorized using imagery. In the current version of the software it is not possible to generate a profile cross-section of the point cloud nor can users view the LAS attributes of individual points by selecting them. Aside from some basic line of sight functionality SOCET GXP does not offer any LiDAR specific analysis tools, but users will find it easy to bring together their point cloud, vector data, and imagery all into a single viewer.

What got me most excited about this release of SOCET GXP were the Automatic Feature Extraction (AFE) capabilities. Previous versions of SOCET GXP included Automatic Terrain Generation (ATG) functionality for producing terrain models from stereo imagery. AFE now works on LiDAR data, allowing one to automatically extract bare-earth surface models without the need for a classified point cloud. It’s important to point out that the AFE capabilities are not limited to LiDAR data and can be used on surface models derived from SOCET GXP’s ATG tools. Although AFE generally performs better when run on LiDAR point clouds, the value of extracting features from surface models derived from stereo imagery should not be overlooked, particularly as commercial UAVs (e.g., Gatewing and Sensefly) have stereo image capabilities.

In the current release AFE extracts only buildings and trees. For the trees the extraction is limited to centroids. For buildings it includes the roof outline, the footprint, the centroid, and the individual facets. AFE provides a whole host of attributes related to the building characteristics such as height, volume, and slope. Providing you have enough of the appropriate SOCET GXP AFE licenses, you can derive buildings and trees from multiple LAS files in parallel. Although the feature extraction process in SOCET GXP is a bit slower than some other commercial solutions I have used on a per file basis, the ability to run processes in parallel means that features can be extracted in hours instead of days, providing you have the requisite number of processing cores.

The AFE interface provides the user with a number of parameters to control, such as the max/min height of buildings. As with any AFE process there is theory and then there is practice, and while I found the AFE documentation provided in SOCET GXP helpful, the best approach to optimizing the settings is one of trial and error. AFE has a nice option that allows the user to specify a polygon subset of LAS tile to run on, which speeds this trial and error process up during the testing phase. I ran AFE on dozens of LAS files from the Berkeley County, WV area and was impressed with the results.

No AFE solution is ever going to be perfect, and in my test case I adjusted the parameters so that there were more errors of commission than omission, with my reasoning being that it is easier to delete features than create them. SOCET GXP typically captured all the buildings for a given area, with limited geometric errors. The most egregious errors in the building detection were near a mining operation and over water, but these were limited and easy to spot.

There were other false positives, and while it is not possible to know the exact cause of these, it could be attributed to the fact that SOCET GXP generates its own bare-earth surface model during the AFE process. There is no option to use the bare-earth classification in the point cloud, an option that I hope to see added in a future release.

Due to the properties of the forest and the LiDAR I was unable to manually identify the precise centroids of individual trees, making a robust evaluation of SOCET GXP’s tree detection AFE approach impossible, but the results "looked good." Errors in tree centroid extraction tended to occur over utility lines and next to buildings. Trees also appeared to be underrepresented in areas of dense canopy. It is important to note that the AFE results are vector features, as SOCET GXP does not have point cloud classification capabilities.

SOCET GXP is unlikely to replace your current LiDAR software suite, but at version 4.0 BAE Systems adds an impressive suite of LiDAR visualization and processing capabilities. With SOCET GXP users can visualize point clouds, generate LiDAR-derived products such as intensity images and bare-earth surface models, and automatically extract building polygons and tree centroids. The AFE capabilities, when combined with SOCET GXP’s existing 2D and 3D feature collection tools, make SOCET GXP an excellent end-to-end solution for keeping spatial databases current. AFE can be used to get the "90% solution" and then the feature collection tools can be used to refine the results.

Jarlath O’Neil-Dunne is the Director of the University of Vermont’s Spatial Analysis Laboratory. He specializes in developing automated techniques to extract information from LiDAR and other types of remotely sensed data.

A 1.791Mb PDF of this article as it appeared in the magazine complete with images is available by clicking HERE

About the Author

Jarlath ONeil-Dunne

Jarlath O'Neil-Dunne ... Jarlath O'Neil-Dunne is a researcher with the University of Vermont's (UVM) Spatial Analysis Laboratory (SAL) and also holds a joint appointment with the USDA Forest Service's Northern Research Station. He has over 15 years experience with GIS and remote sensing and is recognized as a leading expert on the design and application of Object-Based Image Analysis Systems (OBIA) for automated land cover mapping. His team at the SAL has generated billions of pixels worth of high-resolution land cover data from a variety of aerial, satellite, and LiDAR sensors in support of urban forestry planning, ecosystem service estimation, and water quality modeling. In addition to his research duties he teaches introductory and advanced courses in GIS and remote sensing using ArcGIS, ERDAS IMAGINE, eCognition, and QT Modeler. He earned a Bachelor of Science in Forestry from the University of New Hampshire, a Masters of Science in Water Resources from the University of Vermont, and certificates in hyperspectral image exploitation and joint GIS operations from the National Geospatial Intelligence College. He is a former officer in the United States Marine Corps where commanded infantry, counter-terrorism, and geospatial intelligence units.
Contact Jarlath Article List Below