International Conference on Innovative Mechanisms for Industry Applications
(ICIMIA 2017)
An overview of Data Fusion Techniques
Dr.V.S.Krushnasamy Pratiksha Rashinkar
Associate Professor Post Graduate Scholar, MECS
Dept. of EIE, DSCE, Bengaluru-78 DSCE, Bengaluru-78
[email protected] [email protected] Abstact- Multi sensor data fusion is a tool used to Hall[2] and Llinas[3] also came up with a
combine the data from various sensors and gives a more definition of data fusion which inferred that this
reliable and accurate output. It is the integration of data technique of data fusion helps combine data from
and knowledge from several sources. This paper has
multiple sensors and various associated database
reviewed the various technologies, the advantages and the
information and get more accuracy and more specific
classification of the data fusion.
inferences which would not be possible using any one
single sensor.
I. INTRODUCTION This technique has been extensively
employed on multisensory environments which aim at
Every person has a different approach or fusing data from various sources. Their applications
perspective towards the same thing, and we cannot range from soil mapping, discovery science, business
predict which person is more correct or more accurate. intelligence, wireless sensor network, biometrics and
Sometimes it so happens that the results/conclusions of even in oceanography[4].
all the individuals put together may fetch us even better Advantage of this multi sensor data fusion
results. Same is the case with Data Fusion technique. over single source data is an improved estimate of
Data fusion is a process of integration of physical phenomenon via redundant observations. For
multiple data and knowledge representing the same real example, a radar provides the ability to accurately
world object into a consistent, accurate and useful determine the aircrafts range, but has limited ability
representation. Process engineers today seek much to determine the angular direction of the aircraft[5].
more precise and reliable level measurement systems By contrast, an infrared imaging sensor can
because of high demands of sophisticated automated accurately determine the angular direction, but is
processing systems, the need for ever tighter process unable to measure ranges. If both these are combines,
control and an increasingly stringent regulatory it provides an improved determination of location
environment. than could be obtained by either of the two
In general all the tasks that demand any type independent sources.
of parameter estimation from multiple sources can Data fusion techniques are drawn from a
benefit from the use of data/information fusion wide range of areas including artificial
methods. The term information fusion and data fusion intelligence[6], pattern recognizaton, statistical
are typically the same, but data fusion refers to ‘raw estimation[7] and other areas.
data’ or data collected directly from the sensors,
whereas the term ‘information fusion’ refers to already II. LITERATURE SURVEY
processed data. The most accepted definition of data
fusion was provided by the Joint Directors of
Laboratory(JDL) workshop[1] as “A multi-level Data fusion first appeared in the literature in
process dealing with the association, correlation, 1960’s as mathematical models for data
combination of data and information from single and manipulation[8]. It was implemented in the US in the
multiple sources to achieve refined position, identify 1970’s in the field of robotics and defence[9]. In the
estimates and complete and timely assessments of mid-1980s, the Joint Directors of Laboratories formed
situations, threats and their significance.” the Data Fusion Subpanel[10] (which later became
978-1-5090-5960-7/17/$31.00 ©2017 IEEE 694
Authorized licensed use limited to: DALIAN MARITIME UNIVERSITY. Downloaded on April 06,2024 at 10:01:45 UTC from IEEE Xplore. Restrictions apply.
International Conference on Innovative Mechanisms for Industry Applications
(ICIMIA 2017)
known as the Data Fusion Group). With the advent of The lowest level of the Data fusion process is the
the World Wide Web, data fusion thus included data, source preprocessing which includes fusion at signal
sensor, and information fusion[11]. The JDL/DFIG[12] and pixels level. This level also includes the
introduced a model of data fusion that divided the information extraction process. In this level the load
various processes. In 1986 the US Department of on the data fusion system is reduced by allocating data
Defence established the data fusion sub-panel of the to appropriate processes. This also helps concentrate
Joint Directors of Laboratories (JDL) to address some on the data most pertinent to the current situation.
of the main issues in data fusion and chart the new field
in an effort to unify the terminology and
B.Level 1- Object Refined
procedures[13].
III.CLASSIFICATION OF DATA FUSION Object refinement employs the processed data
from the level 0 to be processed by either association,
The Data fusion can be classified into 3 basic clustering or grouping techniques, state estimation etc.
types, according to the requirements of the applications
Level 1 performs these key features
1.JDL data fusion classification where the data fusion
i.A set of units and coordinates is obtained from its
levels are defined by the JDL as level 0, level 1, level 2,
level 3 and level 4. sensor data
2.Dasarathy’s classification systems based on the ii.It gets an object’s position, kinematics and attributes
input/output data types and their nature. iii.It applies data to particular objects to allow the
3.Classification based on abstraction levels of data such statistical estimation of it.
as fusion-feature, decision and hybrid. iv.It refines the estimation of an object’s identity or
Each of these classifications are discussed in detail classification.
below,
C.Level 2-Situation assessment
1. JDL Data Fusion This level focuses on a higher level of inference
the level 1. Here the analysis is performed based on
environmental information about terrain, hydrology,
This classification is most popular conceptual surrounding media at interpreting data, analogous to
model in the data fusion community. It was originally how a human might have interpreted the meaning of
proposed by the JDL and the American department of sensor data.
Defence. These organisations classified the data fusion
process in five processing levels as shown in figure 1,
an associated database, and an information bus that D.Level 3-Impact assessment
connects the five components. This level, it projects the current situation into
future to draw inferences about enemy threats, friendly
and enemy vulnerabilities and any opportunities for
operations.This not only focuses on the possible
engagement outcomes, but also decides or estimates
enemy’s intend based on knowledge about level of
training, political environment and current situation.
Figure 1: JDL Data fusion framework E.Level 4-Process refinement
Level 4 can be considered as a meta-process ie. A
process concerned about other processes, because it
The 5 levels of data processing are defined as follows:
improves the process from level 0 to level 3 by
A.Level 0- Source preprocessing providing resource and sensor management. The aim
978-1-5090-5960-7/17/$31.00 ©2017 IEEE 695
Authorized licensed use limited to: DALIAN MARITIME UNIVERSITY. Downloaded on April 06,2024 at 10:01:45 UTC from IEEE Xplore. Restrictions apply.
International Conference on Innovative Mechanisms for Industry Applications
(ICIMIA 2017)
of this is to achieve efficient resource management 3. ABSTRACTION LEVELS
while accounting for the task priorities, scheduling and
the control of available resources.
A.Early fusion
Here, in early fusion, the various features are
2. Darasathy’s Classification extracted from input data and are combined to produce
a new raw data which is now used as a new input to
Another well known data fusion model is shown in perform better further analysis. This new input is
figure 2 which was proposed by Dasarathy is composed expected to be more informative then the inputs
of the following 5 major categories based on initially taken, as these are processed by compression
input/output data types and their nature. techniques. But increase in the features( especially
from different modalties) becomes complex to handle.
A.Data in-Feature out (DAI-FEO)
B.Late fusion
At this level, the data fusion process employs raw data
from the sources to extract features or characteristics The objective of late fusion is to obtain limited
that describe an entity in the environment. relevant features. This level of fusion allows features
from different representations to be combined in the
same format of representation. Late fusion allows for
more and better scalability and flexibility.
C.Hybrid fusion
Hybrid fusion combines the advantages of both
the early fusion and late fusion. Hence many
multimedia analysis problems are chosen to be solved
using this hybrid fusion.
Table 1: Classification based on Abstraction level
Figure 2: Dasarathy’s input output model
B.Feature in-Feature out (FEI-FEO)
Here, both the input and output of the data fusion
process are features and hence, the data fusion process
addresses a set of features which improves, refine or
obtain new features.
C.Feature in- Decision out (FEI-DEO)
In this level a set of features as input is provided and a
set of decisions is obtained as output. All those systems
that perform a decision based on a sensor’s input fall IV. CONCLUSION
into this category. The data fusion community is rapidly evolving.
D.Decision in- Decision out (DEI-DEO) Significant investments in Department of Defence
(DoD) applications, rapid evolution of
Also called as decision fusion it fuses the input
microprocessors, advanced sensors, and new
decisions to obtain better or new decisions.
techniques have led to new capabilities to combine
data from multiple sensors for improved inferences.
Implementation of such systems requires an
understanding of basic terminology, data fusion
978-1-5090-5960-7/17/$31.00 ©2017 IEEE 696
Authorized licensed use limited to: DALIAN MARITIME UNIVERSITY. Downloaded on April 06,2024 at 10:01:45 UTC from IEEE Xplore. Restrictions apply.
International Conference on Innovative Mechanisms for Industry Applications
(ICIMIA 2017)
processing models, and its classification. This paper is
intended to provide an introduction to these areas as a
basis for further study and research.
REFERENCES
[1] E. Waltz, “Data fusion for C3I: A tutorial,” in Command, Control,
Communications Intelligence (C3I) Handbook. Palo Alto, CA: EW
Communications, 1986, pp. 217–226.
[2] D. Hall, Mathematical Techniques in Multisensor Data Fusion.
Boston, MA: Artech House, 1992.
[3] J. Llinas and E. Waltz, Multisensor Data Fusion. Boston, MA:
Artech House, 1990.
[4] N. Rao “Multisensor fusion under unknown distributions: Finite
sample performance guarantees”. Multisensor Fusion, Kluwer
Academic Publishers, 2002.
[5] E. L. Zaveleta and E. J. Smith, “Goddard trajector determination
systems user’s guide,” Tech. Rep. CSC/SD-7S/6005 Computer Sci.
Corp., Apr. 1975.
[6] R. Mahler, “A unified foundation for data fusion,” in Proc. 1994
Data Fusion Syst. Conf., Applied Phys. Lab., Johns Hopkins Univ.,
June 1987.
[7] I. A. Goodman, “A general theory for the fusion of data,” in Proc.
1987 Joint Services Data Fusion Symp., Applied Phys. Lab., Johns
Hopkins Univ., June 1987. [8]“Modeling and fusing of linguistic,
probablistic, and/or conditional and unconditional forms,” Invited
Tutorial, in Proc. 5th Joint Service Data Fusion Symp., part 2, Oct.
1991.
[9] I. R. Goodman and H. T. Nguyen, Uncertainty Models for
Knowledge-Based Systems. Amsterdam: North-Holland, 1985.
[10] L. I. Perlovsly and J. V. Jaskolski, Concurrent Adaptive
Classification and Tracking, Low Frequency Active Techn. Rev.,
Office of Naval Technol., Washington, DC.
[11] E. Waltz, “The principles and practice of image and spatial data
fusion,” in Proc. 8th Natl. Data Fusion Conf., Dallas, TX, 15–17 Mar.
1995.
[12] E. Waltz, “Data Fusion for C3I: A Tutorial,” Command, Control,
Communications Intelligence (C31) Handbook, EW Communications,
Inc., Palo Alto, CA, pp. 217-226, 1986.
[13] J. Llinas, E. Waltz, Multisensor Data Fusion, Artech House, Inc.,
1990. [3] D. Hall, Mathematical Techniques in Multisensor Data
Fusion, Artech House, Inc., 1992.
[14] L.A. Klein, Sensor and Data Fusion Concepts and Applications,
SPIE Optical Engineering Press, Tutorial Texts, Vol. 14, 1993.
978-1-5090-5960-7/17/$31.00 ©2017 IEEE 697
Authorized licensed use limited to: DALIAN MARITIME UNIVERSITY. Downloaded on April 06,2024 at 10:01:45 UTC from IEEE Xplore. Restrictions apply.
