CC BY
107
УДК 528.71
Александер Вигерт, Михель Грубер Vexcel Imaging GmbH 8010 Грац, Австрия
Email: alwieche, michgrub}@microsoft.com
BING КАРТЫ В КАЧЕСТВЕ ОСНОВЫ ДЛЯ СОЗДАНИЯ ЦИФРОВОЙ МОДЕЛИ ЗЕМЛИ ПО ДАННЫМ ЦИФРОВЫХ АЭРОФОТОКАМЕР ULTRACAM
Alexander Wiechert, Michael Gruber Vexcel Imaging GmbH 8010 Graz, Austria
Email: alwieche, michgrub}@microsoft.com BING MAPS - MAPPING THE WORLD IN 3D BY ULTRACAM ABSTRACT
This paper describes the BING maps project carried out since 2005 by Microsoft. The history is described as well as the underlying technology and the most recent Global Ortho project. Also the digital aerial frame camera UltraCamG is described. This camera has been developed specifically by Vexel Imaging for Microsoft for nation-wide mapping to carry out the Global Ortho project.
Key words: BING maps, Global Ortho, UltraCam, UltraMap, digital camera, aerial camera, remote sensing, digital photogrammetry.
INTRODUCTION
In the last years, global player such as Yahoo, Microsoft or Google entered into the mapping community and invest heavily into the development of internet based mapping services for consumer and business. Certain dataset are used such as street maps, images and other databases to enable services such as street view, aerial view, and search for driving or walking directions, display of traffic information or search for people, business and locations. Also the platforms offer interfaces such as AJAX or Silverlight which enable the user to build own applications and business on top of the platform by using data and features from the platform. Prominent examples of commercial applications are the use of these internet mapping services by real estate companies or for tourism or for location based advertising.
VIRTUAL EARTH
The processor of BING maps was called Virtual Earth. This project has been launched by Bill Gates during his speech in London at his 50 th birthday. The vision
Bill shared was impressing: to map 3000 cities in five years with a fully automated workflow and a cost reduction of 90% over current existing methodologies.
To enable this vision, in 2006 Microsoft purchased Vexcel Imaging GmbH, Austria, a leading manufacturer of digital aerial frame cameras. Since then, Vexcel Imaging became a research and development center for BING maps in addition to its commercial aerial camera business. Vexcel Imaging has developed impressive software which allowed Microsoft to process 3D city models in a highly automated workflow. The underlying methodology is based on multi-ray photogrammetry.
Figure 1: Multi-ray photogrammetry is the underlying methodology for Microsoft 3D city model generation. Specific flight pattern ensures redundancy and avoids
occlusions.
UltraCam images are the sole source for the determination of the vectorized 3D city model. Important for the automated processing are some key image parameters such as redundancy, geometric accuracy and radiometric dynamic. Redundancy has been achieved by a specific flight pattern with 80% forward and 60% sidelap. Due to the leading geometric accuracy and radiometric dynamic of the UltraCam images, it was possibly to process surface models (DSM) and terrain models (DTM) in a fully automated workflow. The DSM has been generated by a dense matching of highly overlapping UltraCam images. That leads to very dense and accurate surface models,
around 10 to 20 times denser than a typical achieved by airborne Lidar sensor systems.
Figure 2: Highly overlapping UltraCam images (left) with 10cm ground sample distance and high resolution DSM (right), processed by dense matching from the images. Point density of the DSM is around 50 points per square meter, accuracy is
better 10 cm.
This DSM is then classified and then filtered automatically into a DTM by using the “knowledge” achieved by the classification.
Figure 3: UltraCam image (left) and classification map (right), derived automatically
from the image.
Figure 4: The DSM (left) which has been gerenated by a dense matchign of the images is filtered into a DTM (right) by a specific algorithm. The classification map
is used during the filter process.
The images, DSM, DTM and classification map are then used to extract a vectored roof polygons of buildings. Then the roof polygon is extruded to the bare earth (represented by the DTM) and leads to a vector description of the building. That building now is available as a vector object, can be textured and attributed and used in the global scene as a building object.
Figure 5: Automatically extracted roof polygon (left), extruded and textured vector building (middle) and global scene with the vectored 3D building model.
The results achieved by this workflow have been impressive in quality, speed and automatization. On the internet, the 3D models enabled rich user experience and have been used in a wide range of applications.
Figure 6: Examples of 3D city models in Virtual Earth, generated by the automated workflow based on multi-ray photogrammetry and UltraCam aerial images.
However, that virtual reality still looks virtual to the end user and more artificial compared to a picture of the real world. Based on the feedback from the user, Microsoft changed the design of the platform.
BING maps (http://maps.bing.com)
The platform BING maps integrates several data sets, including vector maps, satellite images, aerial nadir images, aerial oblique images and the virtual reality 3D layer.
Figure 7: BING maps data sources, such as vector maps (upper left, aerial nadir images (upper right), aerial oblique images (lower left) and virtual reality 3D layer
(lower right)
With the growing availability of the oblique images and the very positive feedback on the aerial oblique images, the virtual reality 3D layer has been deactivated in 2010 and replaced by an image based synth view of oblique images.
Figure 8: Virtual reality 3D layer (left) versus oblique image based 3D look and feel
(right)
This gives an impressive look and feel based on real images, significantly less artificial than the virtual reality 3D layer. Also, the image based representation allows a smooth transition to the street side level. However, even if this approach sounds purely image based, there is still photogrammetry working the background. To allow a smooth transition (moves, rotations, zooms), knowledge of the underlying 3D structure is required, however in a less detailed representation compared to the virtual reality 3D layer. So, there is still a native 3D structure behind the images. This 3D block structure is derived out of the aerial UltraCam images by the process which has been discussed in the previous chapter.
GLOBAL ORTHO
A big challenge of a world-wide mapping project is the consistency and age of data. Even in well developed areas, there is inconsistency in quality, resolution and age of images.
—up. «SHU jiilHiiM
Figure 9: Marin, Illinois, USA - an example of inconsistent color, resolution and age.
To address this, Microsoft has launched in co-operation with DigitalGlobe a project called Global Ortho. Within 24 months, the whole of USA and Western Europe will be captured by aerial images with a 30cm resolution (15cm in some urban areas).
Figure 10: Global Ortho coverage. USA (left) and Europe (right). Color represent priorities, ranging from urban orange) to country side (white).
The coverage of the Global Ortho project is 8.1 million square kilometers in the USA and additional 2,4 million square kilometer in Western Europe. These areas will be mapped by a set of specific cameras called UltraCamG, operated by selected flyers for Microsoft.
ULTRACAM
UltraCam is a leading aerial camera series with more than 205 cameras sold world-wide, developed by Vexcel Imaging GmbH, Austria, for the commercial remote sensing market. The cameras are designed to serve photogrammetric applications with typical image resolutions ranging from 50cm to 2.5cm. The camera series has some outstanding features such as the syntopic exposure which leads to parlay free images, superior monolithic image geometry achieved by monolithic stitching, superior image radiometry of 7600 grey values, achieved by monolithic radiometry and a highly automated, very fast 16 bit image processing workflow with automated features such as distributed processing, project based color balancing and embedded aerotriangulation.
Figure 11: UltraCam camera series - UltraCamXp Wide Angle (left), UltraCamXp
(middle), UltraCamLp (right).
The first camera was the UltraCamD which has been announced 2003. Since then, Vexcl Imaging, has continuously turned the latest available technology into an innovative series of cameras. Vexcel Imaging constantly updated stability, reliability, accuracy and efficiency of the camera system to the then current available technological possibilities to maximize the benefit of the UltraCam customer.
Digital aerial cameras have basically replaced analog cameras for all kind of applications. Since the first digital cameras came to the market, a constant increase of frame size or more generic: an increase of the amount of pixel across the flight strip took place. The key parameter for the collection efficiency of a digital aerial camera is the number of pixel across flight strip. This directly impacts the number of flight lines required to map a certain area. That development was driven by the need of increased flight efficiency to minimize flight costs, minimize flying time and to minimize project risk. The number of pixels along the flight line has no or only very little impact on the collection efficiency because this can always be offset by a fast frame rate and an automated processing workflow.
ljI feV J nS/J' • !Hri
Figure 12: Evolution of the UltraCam frame size. Ranging from UltraCamD (2004, left) to UltraCamX (2006, middle) and to UltraCamXp/UltraCamXpWA (2008/2009,
right).
The current UltraCam models are UltraCamLp with a footprint of 11700 pixels across the flight strip, UltraCamXp and its wide angle version UltraCamXp Wide Angle with 17310 pixels across the flight strip. The cameras are commercially available and successfully used world-wide by a wide range of organizations.
To enable the demanding and huge Global Ortho, Vexcel Imaging GmbH has developed a specific camera called UltraCamG. This camera is especially designed for nation-wide ortho mapping. Key parameters are an impressive footprint of 28000 pixels across the flight strip and a reversed pan-sharpening ration of 3:1 between the Pan and the color channels. Thus, the camera collects high resolution color and lower resolution Pan image information. This is reversed to a standard photogrammetric camera which usually collects high resolution Pan for the photogrammetric applications and lower resolution color for the mapping applications. The Pan cannel of the UCG is not designed for high accuracy photogrammetric applications; it is designed to enable the processing of a DTM suitable to orthorectify the 30cm color images. This makes the camera and the Global Ortho project independent from any other data sources such as airborne Lidar data. The UCG collects the high resolution images, it collects the lower resolution Pan images which are used to process a DTM automatically based on the above described methodology, and it also collects lower resolution NIR for classification.
Figure 13: UltraCamG camera system (left) and a view on the UltraCamG sensor
head and lens system (right)
The pan channel is captured at a large frame with lower resolution and high forward overlap of 85% to ensure automated DSM and DTM production. The color channel has a smaller dimension in flight direction and is exposed with a much higher frame rate and a forward overlap of 20% to ensure coverage. The NIR channel covers only the center part and has full coverage due to suitable forward and sidelap.
Figure 14: UltraCamG channels: low resolution PAN (left), covering the area with 85% frontlap; high resolution color (middle), covering the area at different frame rate with 20% frontlap; low resolution NIR (right) covering the center but area coverage
is ensured by frontlap and sidelap.
Operationally, the camera is designed to be operated at higher altitudes (30cm GSD is being achieved at around 5000m flight altitude and results in an almost 9km wide strip) and very fast air speed. This leads to superior mapping efficiency above all existing camera systems.
Figure 15: UltraCamG frame, Rome, Italy. Shows is oen full frane PAN and one RGB strip. The detail shows the Vatikan. Image resolution is 30cm.
The examples show here give an impression about the outstanding image quality achieved by the UltraCamG. The UltraCamG enables not only Microsoft to map the whole of USA and Western Europe in roughly 24 months only with an impressive image quality, consistency and timeliness. Furthermore, the camera is an excellent example of the leading UltraCam camera design. That design allows developing leading commercial photogrammetric cameras as well outstanding cameras targeting specific applications.
Figure 16: UltraCamG color image example, Frankfurt, Germany. Image resolution is
30cm ground sample distance.
References
Ladstaedter et al, 2010: Monolithic Stitching: One Sensor Geometry For Multiple Sensor Cameras, Proceedings of the American Society for Photogrammetry & Remote Sensing, 26-30 April, 2010, San Diego, CA.
Gruber, M. & Wiechert, A., 2009: New digital aerial cameras by Vexcel Imaging / Microsoft, Proceedings of the Remote Sensing and Photogrammetry Society (RSPSoc) Annual Conference 2009, 8-11. 9. 2009, Leicester, UK.
Mansholt, U., Ladstadter, R. (2008): Geometric analysis of Vexcel Imaging UltraCamX test flights, Proceedings of
the XXI ISPRS Congress, 3-11 July 2008, Bejing, China.
Gruber, M., 2007: UltraCamX, the new digital aerial camera system by
Microsoft Photogrammetry, Proceedings of the Photogrammetric Week 2007, Stuttgart, DE.
Leberl, F. et al. 2003: The UltraCam Large Format Aerial Digital Camera
System, Proceedings of the American Society for Photogrammetry & Remote Sensing, 5-9 May, 2003, Anchorage, AL.
© A. Wiechert, M. Gruber, 2011
