Data fusion is one the fundamental elements of modern tracking techniques, utilising information from a variety of sources and combining the information in a way that meets the desired application constraints and objectives.

Data fusion is a framework describing the process of combining data originating from different sources. The objective of data fusion is maximization of useful information, such that the fused information provides a more detailed representation with less uncertainty than that obtained from individual sources. While producing more valuable information, the fusion process may also allow for a more efficient representation of the data. Another by-product of information fusion may be the observation of higher-order relationships between respective entities.

The selected method for performing the data combination will depend on the original data format produced by the various sensor types. Data fusion is, in general, conducted using one of the following frameworks:

" Pixel level fusion

" Feature level fusion

" High-level data fusion

Pixel level fusion describes the combination of multiple images into a single image, where raw data is robustly and redundantly merged. Each location in the resulting image is an algorithmic combination of the vector of measurements from each of the sensors.

Feature level fusion refers to the extraction of features from each of the sensor data. Registration of detected features is performed for regions of interest or image segments containing more than one pixel. A detection/classification algorithm can then be applied on the combined feature vector.

High-level data fusion or decision fusion occurs where sensor data, with or without pre-processing, is combined with other data or a priori knowledge. Each sensor makes an independent decision based on its own observations and passes these decisions to a central fusion module where a global decision is made. Alternatively, in a decentralized multi-sensor system each node functions performs data fusion based on local observations and the information communicated from neighbouring nodes.

Shown below (Figure 1) are some of the techniques applied to the various elements of the data fusion process.

Figure 1 (Source: http://www.eng.man.ac.uk/mech/merg/Research/datafusion.org.uk/whatis.html )

• Maximization of useful information
• More reliable information than possible from individual sources
• More efficient data and information representation
• Detection of higher-order relationships between different dedicated sensor types

• Differences in field of view, in sensor orientation (e.g. forward looking, downward looking), in resolution and in data format require advanced pre-processing techniques to enable accurate and reliable fusion of data from different sources.
• In the case of high level fusion, the sensors may have very different data characteristics. Consequently, a detailed understanding of the individual data types and how they relate is required for appropriate implementation of decision making systems

Practical applications of data fusion have necessarily been those areas in which the required output of an analysis may not be measured directly.

• This is particularly important such as:
• Medical imaging
• Non-destructive testing
• Remote sensing applications such as target identification and tracking
• Condition monitoring for the detection of faults and degradation of machinery
• Landmine detection

http://www.inforfusion.org/  

http://www.data-fusion.org/index.php  

http://www.eng.man.ac.uk/mech/merg/Research/research1.HTML  

http://www.elsevier.com/locate/inffus  

http://www.acfr.usyd.edu.au/projects/research/Systems%20of%20Systems/Decentralised%20Data%20Fusion/index.html

1. Wald L., Some terms of reference in data fusion. IEEE Transactions on Geosciences and Remote Sensing , 37, 3, 1190-1193, 1999.
2. F. Cremer, K. Schutte, J.G.M. Schavemaker and E. den Breejen. A comparison of decision-level sensor-fusion methods for anti-personnel landmine detection . Information Fusion , 2 (2001)
3. R.J. Stanley, P.D. Gader and K.C. Ho. Feature and decision level sensor fusion of electromagnetic induction and ground penetrating radar sensors for landmine detection with hand-held units . Information Fusion , 3 (2002)
4. Zaatri and M. Oussalah. Integration and design of multi-modal interfaces for supervisory control systems . Information Fusion , 4 (2003) 135-150
5. Xubo B. Song, Yaser Abu-Mostafa, Joseph Sill, Harvey Kasdan and Misha Pavel Robust image recognition by fusion of contextual information . Information Fusion , 3 (2002) 277-287
6. D. Rajan and S. Chaudhuri. Data fusion techniques for super-resolution imaging . Information Fusion , 3 (2002)
7. Belur V. Dasarathy. Information fusion as a tool in condition monitoring . Information Fusion , 4 (2003) 71-73
8. G. Simone, A. Farina, F.C. Morabito, S.B. Serpico and L. Bruzzone. Image fusion techniques for remote sensing applications . Information Fusion , 3 (2002)
9. Belur V. Dasarathy. Information fusion, data mining, and knowledge discovery . Information Fusion , 4 (2003) 1-1
10. Sun H, et al, 1994, Study on an algorithm of multisensor data fusion, Proc. IEEE Proceedings of the National Aerospace and Electronics Conference, Vol.1, pp.239-245
11. Korona Z, Kokar M M, 1996, Model-based fusion for multisensor target recognition, Proc. SPIE, Vol.2755, pp.178-189,
12. Thomas J H, Dubuisson B, 1996, Diagnostic method using wavelets networks application to engine knock detection, Proc. IEEE International Conference on Systems, Man and Cybernetics, Vol.1, pp.244-249,
13. Lou K N, Lin C J, 1997, Intelligent sensor fusion system for tool monitoring on a machining centre, International Journal of Advanced Manufacturing Technology, Vol.13, No.8, pp.556-565
14. Park S, Lee C S G, 1993, Fusion-based sensor fault detection, Proc. 1993 IEEE Int. Symp. Intelligent Control, pp.156-161
15. Grime S, Durrant-Whyte H F, 1994, Data fusion in decentralized sensor networks, Control Engineering Practice, Vol.2, No.5, pp.849-863