A Few Examples Of Systems Exploiting Information Fusion

Information fusion is routinely exploited in daily applications. Examples of such systems are given.

Data fusion is exploited by a large number of biological systems.

An illustration is given by the human system, which calls upon its different senses to perceive its environment. Human sensors acquire information on sight, smell, touch, hearing, and taste. The acquired data are processed within the brain. To do so, the brain will use other sources of information: its memory, its experience and its a priori knowledge. Calling upon its reasoning capabilities, the brain "fuses" all this available information to perform deductions, to produce a representation of the environment and to order action.

This example also illustrates how much data fusion is at the crossings of several scientific domains. Here neuroscience, sciences of cognition, and medicine are at stake.

Data fusion is not limited to biology. It originates in Defense activities, and such applications are still very vivid. Nevertheless, fusion applies to many other domains.

Examples are numerous in transportation, and especially in civil aviation (aids in aircraft, air traffic control, landing aids) and motorways management. Large research efforts are devoted to intelligent car traffic, where each car embark a set of sensors and fusion capabilities, in order to best co-operate with other vehicles and the environment itself. Navigation / positioning is a service routinely offered today. An efficient service calls upon the fusion (often called hybridization in this domain) of several sources: fleets of orbiting satellites and ground systems. Telephone is another example, where several resources must be used through complex fusion systems to make a phone call: transponders aboard geostationnary or low Earth orbiting satellites and terrestrial networks. Robotics calls upon data fusion for 3-D vision and displacement in hostile environment, monitoring, inspection and maintenance of pieces of equipment.

The exploitation of satellite images and more generally of observations of the Earth and our environment is presently one of the most productive in data fusion. Observation of the Earth is performed by means of satellites, planes, ships, and ground-based instruments. It results into a great variety of measurements, partly redundant, partly complementary. There are very few domains, where such a diversity is present and this makes Earth observation so fascinating. The availability of so many types of information constitutes a tremendous field of investigation for mathematicians. This interest is enhanced by the challenge of correctly modeling natural landscapes and outdoor scenes, which are usually more difficult than indoor scenes. The research in this field is backed up by the present political interest in environment and global changes.

These measurements in Earth observation may be punctual and time-integrated, bi-dimensional and instantaneous (images), vertical profiles with time-integration or not, three-dimensional information (oceanic / atmospheric profiler / sounder at ground level, or satellite-borne, or ship-borne). Adding the large amount of archives and numerical models representing the geophysical / biological processes, one should conclude that the quantity of information available to describe and model the Earth and our environment increases rapidly. Data fusion is a subject becoming increasingly relevant because it efficiently helps scientists to extract increasingly precise and relevant knowledge from the available information.

The set of sensors for Earth observation is extremely various. The spectrum of their characteristics is very large, with respect to spatial and temporal scales, spatial and temporal sampling and means of acquisition. Such diversity is a tremendous source of practical problems, whose resolutions lie upon a good understanding and modeling of more fundamental questions. For example, what are the links between temperature measurements made at ground level using a thermometer and integrated over an hour, and the instantaneous measurements of the same temperature but made using a satellite-borne radiometer sensing the radiation emitted by a surface of several square kilometers? Data fusion is here at the crossings of the physics of the measurements, Earth sciences and sciences of information and communication. These crossings offer many opportunities and benefits to the progresses in data fusion.

Weather forecasting fully illustrates data fusion in environment. It is one of the most sophisticated fusion systems nowadays and is performed several times a day for the whole planet. It calls upon sensors, signal processing, artificial intelligence and complex modeling of physics and chemistry and the atmosphere, oceans and land. There are processing issues, topological issues (the distribution of sensors in 3-D space and time) and communication challenges.

Meteorological satellites are orbiting the Earth, in a geostationnary orbit or in a near-polar one. They are equipped with sensors providing sets of measurements on the 3-D properties of the atmosphere and on the characteristics of the surface of the ground and the ocean. Balloons and planes operate at lower altitudes. Tens of thousands of ground stations are distributed in the world. They measure the basic weather parameters, such as air temperature and pressure, wind, cloudiness, rainfall, and more for some of them. Ground radars follow storms and rain cells. At sea, ships and automated buoys provide weather parameters and measurements of the sea surface, such as temperature and waves. All these measurements are processed to extract geophysical parameters of interest, and transmitted by means of specialized communication networks. Then in numerical weather prediction centers, numerical models through data assimilation techniques ingest this wealth of information, together with weather prediction of the previous instants. They produce weather forecast that are used by professionals and are also presented on TV news and other media.

31-08-2001 - Copyright L. Wald, Armines / Ecole des Mines de Paris