Increased Image Collection Opportunities
DigitalGlobe’s WorldView-1 Satellite

The successful operation of DigitalGlobe WorldView-1 has created a new milestone for high resolution satellites. The high resolution sensor and the improvement to the onboard equipment have advanced mapping using high resolution images even without the need for ground control points.

By Philip Cheng and Chuck Chaapel

Figure 1: QuickBird image.

Figure 2: WorldView-1 image.

 

WorldView-1, built by Ball Aerospace and Technologies Corporation with the imaging sensor provided by ITT Corporation, is a high-capacity, panchromatic imaging system featu­ring half-meter resolution imagery. With a nominal swath width of 17.6 km at nadir and an average revisit time of 1.7 days, WorldView-1 is capable of collecting up to 750,000 square kilometers (290,000 square miles) per day of half-meter imagery.  Frequent revisits will increase image collection opportunities, enhance change detection applications and enable accurate map updates. The satellite is capable of collecting, storing and downlinking more frequently updated global imagery pro­ducts than any other commercial imaging satellite in orbit, allowing for expedited image capture, processing and delivery to customers where speed is a driving factor. WorldView-1 is equipped with state-of-the-art geo-location accuracy capability and exhibits unprecedented agility with rapid targeting and efficient in-track stereo collection.

WorldView-1 is part of the National Geospatial-Intelligence Agency’s (NGA) NextView program, and was partially financed through an agreement with the NGA. The majority of the imagery captured by WorldView-1 for the NGA will also be available for distribution through DigitalGlobe’s Image Library. Additionally, WorldView-1 immediately frees capacity on DigitalGlobe’s QuickBird satellite to meet the growing commercial demand for multi-spectral geospatial imagery. It is likely that in the future NGA will decide to move some of their orders from QuickBird to WorldView-1, which will allow for more tasking capacity on QuickBird.

WorldView-1 is the first of two new next-generation satellites DigitalGlobe plans to launch in the near future.  In late 2008, Ball Aerospace and Technologies Corporation and ITT Corporation will complete WorldView-2, bringing the total number of satellites DigitalGlobe has in orbit to three; thus enabling the company to offer a constellation of spacecraft that will provide the highest collection capacity – more than 1 million square kilometers per day – of high-resolution Earth imagery directly to customers around the world. Additionally, WorldView-2 will provide eight bands of multi-spectral for life-like true color imagery and greater spectral applications in the mapping and monitoring

markets. This article will examine different areas of the WorldView-1 satellite image data. Firstly, the image data will be compared with the QuickBird satellite data of the same area.  Secondly, the geometric correction method and accuracy of the WorldView-1 data will be examined. Given that the WorldView-1 is equipped with state-of-the-art geo-location accuracy, it would be useful to find out the geometric model accuracy of the WorldView-1 data with and without ground control points (GCPs). Lastly, we will test the pan-sharpening of WorldView-1 data using QuickBird multispectral data. 

Figure 3: WorldView-1 image of Spokane.

Figure 4: WorldView-1 image of Castle Rock.

Figure 5: WorldView-1 image of Phoenix.

WorldView-1 Data

Similar to the QuickBird satellite data, WorldView-1 data is distributed in five diffe­rent levels, i.e., Basic 1B, Basic Stereo Pairs, Standard 2A, Ortho-Ready Standard OR2A, and Orthorectified. For custom orthorectification the Standard 2A and Orthorectified pro­ducts are not recommended. Standard 2A product is not re­commended because of the coarse DEM correction already applied to the image data. Basic Imagery products are the least processed of the WorldView-1 Imagery Products. Each strip in a Basic Imagery order is processed individually and therefore, multi-strip Basic Imagery pro­ducts are not mosaicked. Basic Imagery pro­ducts are radiometrically corrected and sensor corrected, but not projected to a plane using a map projection or datum. The sensor correction blends all pixels from all detectors into the synthetic array to form a single image. The resulting GSD varies over the entire product because the attitude and ephemeris slowly change du­ring the imaging process. 

Basic Stereo Pairs are supplied as two full scenes (490 km2 ) with 90% overlap, designed for the creation of digital elevation models (DEMs) and derived GCPs.

Ortho-Ready Standard Imagery has no topographic relief applied, making it suitable for custom orthorectification. Ortho-Ready Standard Imagery is projected to an average elevation, either calculated from a terrain elevation model or supplied by the customer. It can be ordered from a minimum of 25 km2 from the Library, or from 64 km2 for new tasking.

QuickBird image and WorldView-1 Image

Figure 6: Mosaicked WorldView-1 image of Phoenix.

In order to leverage the WorldView-1 images for applications such as GIS, it is necessary to orthorectify the images.  A geometric model, GCPs and DEMs are required. The RPC model has been the most popular method in orthorectifying high resolution images.   More details about the RPC model can be found in the paper written by Grodecki and Dial (Block Adjustment of High-Resolution Satellite Images Described by Rational Functions - PE &RS January, 2003).  

The latest version of PCI Geomatics’ OrthoEngine software was used for this tes­ting.  This software supports reading of the data, manual or automatic GCP/tie point (TP) collection, geometric modeling of different satellites using Toutin’s rigorous model or the RPC model, automatic DEM generation and editing, orthorectification, and either manual or automatic mosaicking. OrthoEngine’s RPC model is based on the block adjustment method developed by Grodecki and Dial and was certified by Space Imaging www.pcigeomatics.com/support_center/tech_papers/rpc_pci_cert.pdf). Since biases or errors still exist in the RPCs, the results can be post-processed with a polynomial adjustment and several accurate GCPs. PCI Geomatics’ OrthoEngine RPC model computes the polynomial adjustment math model for each image.

Where A0, AS, AL, ASL …  and B0, BS, BL, BSL …  are the image adjustment parameters, Line and Sample are the line and sample coordinates of an image, and ∆P and ∆R are the adjustable functions expressing the differences between the measured and the nominal line and sample coordinates. The OrthoEngine software supports zero, first and second order RPC polynomial adjustments. It is recommended to use zero order for IKONOS satellite data, first order for QUICKBIRD satellite data, and second order for IRS AWiFS satellite data. One of the purposes of this article is to determine which order of RPC polynomial adjustment would be sui­table for WorldView-1 satellite RPC data.

Although the RPC model only requires a small number of GCPs and TPs, high accuracy may not be achieved if the GCPs are not well distributed within the block. To improve the relative accuracy, a DEM can be used if it is

available. During each bundle adjustment iteration, the computed elevation of each tie point can be replaced by the elevation at the computed TP X and Y coordinates from the DEM, similar to the results of changing the planimetric TPs into altimetric points. This method helps improve the relative accuracies between the ortho images, which helps to minimize differences during the mosaicking

process. This option is available within the OrthoEngine software.

Spokane Test Results

To generate an orthorectified WorldView-1 image with coverage of about 18km by 18km (2.7 Gigabytes) requires approximately 18 minutes on a Pentium IV 3.0 GHz machine running Windows XP. 

Table 1: Comparisons of Basic 1B and ORZA RPC results of Spokane Meters

Table 2: Comparisons of ORZA RPC results of Castle Rock in meters.

Table 3: Comparisons of ORZA RPC results of Phoenix in meters.

Table 4: Comparisons of Phoenix ORZA RPC block results in meters.

Castle Rock Test Results

Phoenix Test Results

Figure 5 shows a WorldView-1 image of Phoenix. A total of 5 ICPs were collected from the OR2A product. The area has an elevation range of 300m to 500m. Similar tests were performed to the data and table 3 shows a summary of the results. The RMS accuracy without using GCPs is within 1.3m and maximum error within 1.6m. The RMS accuracy improved to within 0.6m when using only 1 GCP.  

A block of 8 images of Phoenix with overlaps were also provided.

The block covers an area of 50km by 44km. A total of 14 ICPs and 16 tie points were collected from the block. The ICPs could only be collected on the left side of the block due to availability. Tie points were collected for the entire block and USGS 30m DEM was used during the RPC bundle adjustment to improve the accuracy. Table 4 shows a summary of the results.  

The RMS accuracy is within 2.1m with maximum error within 3.2m when no GCPs were used. When 4 corner GCPs were used, the RMS accuracy was improved to within 1m with maximum error within 1.4m. Figure 6 shows the mosaicked result of the images using PCI Geomatics’ OrthoEngine automatic mosaicking and color balancing software.  The red color represents the locations of the GCPs. To check the accuracy for the right side of the mosaic where GCPs were not used, USGS 0.25m color photos were obtained from the USGS web site. Figure 7 shows the WorldView-1 image together with the USGS color photo. The two images overlap nicely with each other.  

Figure 7: WorldView-1 panchromatic image together with USGS color photo

Pan-sharpening

The availability of a 1m panchromatic band, in conjunction with 4m multispectral bands, provides the opportunity to create a 1m multispectral pan-sharpened image by fusing these images. The concept of fusion for multispectral images is not new.  Landsat MSS data (bands 4, 6 and 7) have been spatially enhanced (from 240m to 80m resolution) by using weighted high-

frequency information from band 5 at 80m resolution. Previous techniques used different weighting coefficients for the panchromatic band and multispectral bands. The RGB-IHS transformation is another common approach, where the high-resolution panchromatic band replaces the intensity channel derived from the lower resolution multispectral channels. Although these alternate techniques yield enhanced imagery that appear to be sharper, they destroy the spectral characteristics of the data. 

Since most earth resource satellites, such as the SPOT, IRS, Landsat 7, IKONOS, and QuickBird, provide both multispectral images at a lower spatial resolution and panchromatic images at a high spatial resolution, it is possible to perform pan-sharpening for both of these images.  However, most of the existing techniques which perform suitably well with medium-resolution images, can hardly satisfy the pan-sharpening of multispectral and panchromatic high resolution images. 

Based on the thorough study and analysis of existing pan-sharpening algorithms and their fusion effects, a new automatic pan-sharpening algorithm has been developed by Dr. Yun Zhang at the University of New Brunswick, in New Brunswick, Canada. This new technique solved the two major problems in pan-sharpening – color distortion and ope­rator dependency. A method based on least squares was employed for a best approximation of the grey level value relationship between the original multispectral, panchromatic, and the pan-sharpened image bands for a best color representation.  A statistical approach was applied to the pan-sharpening process for standardizing and automa­ting the pan-sharpening process. The new algorithm is available within the PCI Geomatics software.

It is possible to perform pan-sharpening of WorldView-1 data using QuickBird multispectral data.  Both data need to be orthorectified first. To test pan-sharpening using QuickBird and WorldView-1 data, a QuickBird multispectral image and a WorldView-1 image of Phoenix were used.  Figures 8, 9 and 10 show the QuickBird multispectral, WorldView-1 panchromatic, and pan-sharpened WorldView-1 using QuickBird multispectral image, respectively. It is also possible to use color air photo to pan-sharpen the WorldView-1 image. Figure 11 and 12 show the USGS color air photo and pan-sharpened WorldView-1 image using USGS color air photo, respectively.

Figure 8: QuickBird multispectral image

Figure 9: WorldView-1 panchromatic image.

Figure 10: Pan-sharpened WorldView-1 using QuickBird multispectral image.

Automated Batch Processing

Since it is possible to generate high accuracy WorldView-1 orthos and color-balanced mosaics within 2m RMS without using GCPs, it is possible to integrate all the processes in a fully automated batch system.   The batch programs required to perform all the steps are available within PCI Geomatics software. It can be run through python or PCI EASI scripts. An automated GCP collection process can be used if higher accuracy is required.   The advantages of automated processing are (1) maximize production, (2) automation of repetitive time-consuming tasks to produce con­­si­stent results, (3) gain operating efficiencies, (4) reduce labor costs, and (5) shorten throughput time for the delivery cycle.   The generation of a large quantity of high accuracy orthos or mosaics, such as a mosaic of an entire country, can be generated easily using the automated system. Multiple computers can also be used to speed up the proces­ses. The fully automated process means that it is easy to generate WorldView-1 orthos/mosaics for quick turnaround. 

Figure 11: USGS color air photo.

Figure 12: Pan-sharpened WorldView-1 using USGS color air photo image.

Conclusions

This article examines different aspects of the WorldView-1 data. The WorldView-1 image quality is better than the QuickBird panchromatic data.

The RPC model can be used as the geometric model to orthorectify WorldView-1 data. It is possible to achieve RPC model accuracy within 2m RMS without using GCPs and within 1m RMS with a

minimum of one GCP for WorldView-1 data. To improve the RPC model accuracy, a zero order polynomial RPC adjustment can be used.

Pansharpening of WorldView-1 data can be performed by using USGS color air photo/QuickBird multispectral and WorldView-1 images to

create a pan-sharpened WorldView-1 image.

Dr. Philip Cheng cheng@pcigeomatics.comis a senior scientist at PCI Geomatics. 

Mr. Chuck Chaapel cchaapel@digitalglobe.comis a senior geospatial engineer at DigitalGlobe.

Have a look at: www.digitalglobe.com

The authors would like to thank DigitalGlobe for providing the test data.