G. G. Wilkinson
Space Applications Institute
Joint Research Centre
European Commission
21020 Ispra (Va.), ITALY.


  1. Definition of the Earth Observation Sector
  2. Current State of the Art Concerning Use of Computer Vision Techniques in Earth Observation
  3. Future Trends in Earth Observation Systems and Potential Impact of Computer Vision Methodologies
  4. References

Definition of the Earth Observation Sector

Earth observation refers to the use of orbiting remote sensing satellite tech- nology to monitor and map the state of the global environment. Satellites are used for studying the oceans, atmosphere, land surface, and biosphere. Considerable emphasis is given currently to monitoring human effects on the global environment and on establishing causal or synergistic links be- tween long term environmental changes. Earth observation is used to gath- er information about natural resources, such as mineral deposits, and to provide timely information on weather patterns, crop state, land use etc.

Apart from a few specialised sounder and scatterometer systems, the ma- jority of satellite-borne sensors are of an imaging type which operate typically in a line-scan or a push-broom mode. Pixel resolutions currently range from several meters to several km. at the Earth's surface.

Imaging sensors now operate in different parts of the electromagnetic spectrum ranging from the optical, through the infra-red and into the microwave region [1]. The majority of optical and infra-red sensors are "passive" (i.e. inert collectors of reflected solar or emitted thermal radiation) -such as the current LANDSAT-5 Thematic Mapper system and the SPOT-3 High Resolution Visible system. The emphasis in the microwave region is for "active" synthetic aperture radar (SAR) systems (e.g. on ERS-2 and RADAR- SAT) which transmit polarised radar pulses and measure the backscatter signal.

Images are usually gathered either from geostationary or sun-synchronous orbits. The former permit images to be regenerated at short time intervals (e.g. every 30 minutes) whilst the latter permit imaging "revisits" to the same location on Earth over timescales ranging from hours to weeks. Some sensors now permit stereo observation (e.g. SPOT). Many applications require fusion of data from different sensor systems (including at multiple resolutions) and also the use of multi-temporal time series.

Earth observation is currently in an expansive phase with new satellites planned for launch over the next 10 years as a result both of new government programmes (e.g. NASA's "Mission to Planet Earth" [2],[3]) and also commercial mapping programmes.

There is a close link between Earth observation and the field of geographical information analysis. Many products of Earth observation are ultimately stored in digital map form in Geographical Information Systems (GIS).

Current State of the Art Concerning Use of Computer Vision Techniques in Earth Observation

Most applications of Earth observation involve the conversion of multi-channel image data into thematic maps via classification procedures. In contrast to most machine vision applications, the required end products are mostly two dimensional maps -i.e. without depth information or derived statistical data. Computer vision techniques have therefore not had much importance in Earth observation to date -except in low-level processing (image filtering, contrast enhancement, edge detection, region segmentation etc). Although there has been interest in using satellite imagery for topographic mapping for which computer vision approaches would be appropriate, it has been generally accepted until now, that the spatial resolution of images currently on offer is not sufficient to identify the majority of useful topographic features [4]. This situation is changing however with imminent resolution increases (see section 3 below).

At present, the use of higher level computer vision techniques in remote sensing is primarily limited to the exploitation of ancillary knowledge and simple shape models for structural interpretation of scenes. A number of good examples are documented in [5]. In most cases, such approaches have been based on expert systems e.g. [6]-[8]. These techniques have mostly been used for the detection of linear features such as roads [9], [10], geological lineaments [11] and hydrological networks [12]. Most of the work in this field has so far been experimental or pre-operational. Interest also exists in the use of image understanding and machine vision techniques within GIS [13] e.g. for the analysis of digitized cadastral maps [14].

Besides use of models and ancillary knowledge for shape detection, much effort has been devoted in the last few years to the improved definition of landscape parcels in images -e.g. via integrated edge detection and region growing segmentation techniques [15], and use of integrated spectral and texture analysis for region characterisation, e.g. [16].

A further use of machine vision techniques is in the derivation of digital terrain models from stereo imagery (e.g. by use of the SPOT satellite [17]). Operational algorithms now exist for stereo satellite image matching enabling the routine derivation of digital terrain models. Such approaches have also been augmented by synthetic aperture radar interferometry techniques in the last few years (e.g. using the ERS-1 and ERS-2 satellites) which also enable surface deformations to be measured (e.g. following earthquakes [18]).

A number of environmental phenomena of interest are dynamic in nature and motion analysis has also recently profitted from computer vision techniques such as optical flow analysis [19]. Also various deformable geometrical models have been used to model the shapes of dynamic phenomena such as oceanic vortices [20].

Future Trends in Earth Observation Systems and Potential Impact of Computer Vision Methodologies

The use of computer vision methodologies in remote sensing is likely to rise significantly in the near future on account of the following trends in Earth observation: (i) The launch of imaging sensors with very high spatial resolution. This will begin to make it possible to use more complex shape models and context information in the analysis of man-made structures. This is likely to revolutionize topographic mapping and urban zone monitoring in particular.

Typical satellites/ sensor systems are:

(ii) The development of multi-view angle sensing -including along-track stereoscopy. This will permit: the use of bi-directional reflectance information to improve environmental monitoring (esp. vegetation state), the construction of accurate digital terrain models, and the derivation of three-dimensional atmospheric information.

Typical satellite / sensor systems are:-

EOS-AM-1 --MISR instrument: 9 separate view angles (launch expected in 1998)

ALOS --AVNIR-2 instrument: 3 view angles in along track direction -vertical and 24 degree inclination fore and aft (launch expected in 2002)

The high resolution imaging satellites CRSS, EarlyBird and QuickBird (see above) will also provide along track stereo viewing capability.


[1] Kramer, H. J. ,1996, Observation of the Earth and Its Environment. Survey of Missions and Sensors, Third Edition (Berlin: Springer-Verlag).

[2] Asar, G. and Greenstone, R. (editors), 1995, Mission to Planet Earth: Earth Observing System Reference Handbook, (Greenbelt MD: NASA God- dard Space Flight Center).

[3] Barnes, W. L. (editor),1993, Sensor Systems for the Early Earth Observing System Platforms, SPIE Proceedings, Volume 1939.

[4] Hartley, W. S., 1991, Topographic mapping and satellite remote sensing: is there an economic link ?, International Journal of Remote Sensing, 12, 9, pp. 1799-1810.

[5] Browning, K. A., Conway, B. J., Muller, J.-P. A. L., and Stanley, D. J. (editors), 1988, Exploiting Remotely Sensed Imagery (London: The Royal So- ciety).

[6] McKeown, D. M., 1987, The role of artificial intelligence in the integration of remotely sensed data with geographic information systems, IEEE Transactions of Geoscience and Remote Sensing, 25, 3, 330-348.

[7] Goodenough, D. G., Goldberg, M., Plunkett, G. and Zelek, J., 1987, An expert system for remote sensing, IEEE Transactions of Geoscience and Remote Sensing, 25, 3, 349-359.

[8] Argialas, D. P. and Harlow, C. A., 1990, Computational image interpretation models: an overview and a perspective, Photogrammetric Engineer- ing and Remote Sensing, 56, 6, 871-886.

[9] Wang, F. and Newkirk, R., 1988, A knowledge-based system for highway network extraction, IEEE Transactions on Geoscience and Remote Sens- ing, 26, 5, 525-531.

[10] Van Cleynenbreugel, J., Fierens, F., Suetens, P., and Oosterlinck, A., 1990, Delineating road structures on satellite imagery by a GIS-guided technique, Photogrammetric Engineering and Remote Sensing, 56, 6, 893-898.

[11] Karnieli, A., Meisels, A., Fisher, L. and Arkin, Y., 1996, Automatic extraction and evaluation of geological linear features from digital remote sensing data using a Hough transform, Photogrammetric Engineering and Remote Sensing, 62, 5, 525-531.

[12] Hadipriono, F. C., Lyon, J. G., Thomas Li, W. H. and Argialas, D., 1990, The development of a knowledge-based expert system for analysis of drainage patterns, Photogrammetric Engineering and Remote Sensing, 56, 6, 905-909.

[13] Gahegan, M. and Flack, J., 1996, A model to support the integration of image understanding techniques within a GIS, Photogrammetric Engineer- ing and Remote Sensing, 62, 5, 483-490.

[14] Lee, L.-H. and Su, T.-T., 1996, Vision-based image processing of digitized cadastral maps, Photogrammetric Engineering and Remote Sensing, 62, 5, 533-538.

[15] Le Moigne, J. and Tilton, J. C., 1995, Refining image segmentation by integration of edge and region data, IEEE Transactions on Geoscience and Remote Sensing, 33, 3, 605-615.

[16] Ryherd, S. and Woodcock, C., 1996, Combining spectral and texture data in the segmentation of remotely sensed images, Photogrammetric En- gineering and Remote Sensing, 62, 2, 181-194.

[17] Muller, J.-P. A. L., Key issues in image understanding in remote sensing, Phil. Trans. Royal Society of London, Series A, 324, 381-395.

[18] Massonet, D. and Adragna, F., 1993, A full-scale validation of radar interferometry with ERS-1: the Landers earthquake, Earth Observation Quarterly, 41, 1-5.

[19] Sun, Y., 1996, Automatic ice motion analysis from ERS-1 SAR images using the optical flow method, International Journal of Remote Sensing, 17, 11, 2059-2087.

[20] Berroir, J. P., Bouzidi, S., Herlin, I. I. and Cohen, I., 1994, Vortex segmentation on satellite oceanographic images, SPIE Proceedings, Volume 2315, pp. 635-645.