With all the recent talk of UltraMap on the UltraCam Blog and with the UltraMap 3.0 Tech Preview webcast coming up quickly, I am reminded that I have been sitting on a paper written by David Day, Director of IT for Keystone Aerial Surveys, about Keystone’s transition in 2010 to the UltraMap Platform and the benefits they have seen as a result. Because David’s write-up is thorough, I will share this in chunks and spread across several blog posts and several days. Remember that the blog is meant to be interactive so be sure to add any comments or questions you may have because I am betting David will be reading and can jump in to reply.
– Jerry Skaw, Sales and Marketing, Microsoft UltraCam Team
Benefits of UltraMap for Keystone Aerial Surveys
In early 2010 Keystone decided to make the switch from using Microsoft’s Office Processing Center (OPC) for processing UltraCam imagery to Microsoft’s UltraMap platform. Keystone had been processing imagery since 2007 using OPC with a custom designed distributed processing system. While UltraMap’s distributed system was reason enough for many users to adopt the UltraMap software platform, the first release did not necessarily provide new functionality for Keystone. The following study is the result of the effort to find other ways that UltraMap can enhance any production workflow for any mapping product in order to justify some of the added costs of converting to the UltraMap platform.
The software that Keystone developed to distribute OPC processing used a SQL server and over 30 processing cores on 12 servers and workstations. An individual image took approximately 5 minutes to process from the raw LVL00 state to LVL02. A job of 2000 images took about 12 hours (or 36 seconds per image) to process from LVL00 to LV02 . The delivery phase, LVL02 to LVL03, was done by setting up the project radiometry in OPC and saving the radiometry to a file. Project wide color balance was limited to OPC’s tools and viewing capabilities and the RAM of the machine setting up the process. The setups were done using the quickviews generated in the LVL02 processing, so no viewing of the final resolution image was possible. The advantages of this system were its speed and flexibility of hardware.
Because the UltraMap framework requires a 64-bit operating system and a large amount of RAM, many of our servers and all of our workstations were not acceptable as UltraMap processing nodes. Thus, Keystone purchased 64-bit operating systems and RAM for their current servers to meet the minimum requirements to allow them to run UltraMap. A replacement plan was developed to create an optimal processing center by acquiring servers that meet the specifications for UltraMap performance. While the older servers were able to run UltraMap, it was not until Keystone purchased new hardware that the UltraMap platform was fully tested. With the new hardware, a single image on a server with 8 cores and 16gb of RAM requires 10 to 15 minutes to process from LVL00 to LVL02. An UltraMap distributed system with six 8 core machines with 6 cores per machine enabled for processing for a total of 36 used cores, a job of 2000 images takes 36 hours (or 68 seconds per image) to process from LVL00 to LVL02. It is the prospect of this investment along with concerns of loss of throughput (compared to the Keystone custom system based on the UltraCam OPC software) that prompted Keystone to study the benefits of the upgrade to UltraMap.
[UltraCam Team Comment: This is an unusually high value and depends on the processing power and I/O speed of the server system. Typical processing durations are around 7 minutes for a single frame on a high-end single computer not using multiple machines for parallel processing.]
UltraMap Features – Monolithic Stitching, Dragon Fly and Camera Operating System (COS) Integration
As is well known, an UltraCamX or Xp LVL02 PAN image is the result of stitching 9 panchromatic images together creating one “virtual” image. In an UltraCamX, the area covered by overlap of the pan images is approximately 7% of the final image and 7% between adjacent images. Point matching and triangulation of these images can be challenging in very unstructured terrain. Monolithic stitching is the term used to describe the technique that uses the lower resolution Red, Green, Blue or NIR imagery to assist in the tie point extraction. This process partly contributes to the increased processing time, but it is very important for desert and water areas. Keystone captured a large dataset several miles out into the Atlantic Ocean, which would not have been possible without the stitching ability.
The LVL00 to LVL02 processing includes DFI file creation (DragonFly files). This allows the user to view the dataset quickly as a whole and then zoom into view the detail of the image. Keystone uses these images in the UltraMap Viewer and, of course, for the AT and Radiometry modules. Keystone had built tools to view the quickviews of the LVL02 imagery for checking, so the UltraMap Viewer only added geospatial functionality for Keystone. This may not be the case for other companies as image checking may only have been done at the LVL03 stage. The Radiometry module, however, did add a great deal of functionality for Keystone. The ability to view the entire project in a geographic view and then quickly zoom into the detail of the image is a major step forward in the ability to color balance imagery correctly. This makes the added time to create LVL02 dragon fly images more acceptable.
At the same time as UltraMap was originally released, a new Camera Operating System (COS) was also released. There were many additional features and enhancements incorporated in that COS release, but those related to improved interaction with UltraMap matter in this context. The ability to perform Project Based Color Balance (PBCB) techniques rely on accurate meta data about the camera’s environment at the time of capture. Data such as time of day, location and date play a factor in sun angle corrections and other PBCB enhancements. Other factors, in some cases undocumented, seem to also affect the geometry of the virtual and pan-sharpened imagery.
[Watch for the next blog post for this topic: Monolithic Stitching & Accuracy Part I]