多传感器数据融合：概念与思想

5星 · 超过95%的资源需积分: 12 34 浏览量更新于2024-07-18 收藏 4.15MB PDF 举报

"Data Fusion- Concepts and Ideas" 是一本由H.B. Mitchell博士编著的专业书籍，主要探讨了多传感器数据融合的理论和概念。这本书是2007年第一版的大幅修订版，增加了新的章节、内容、例子和参考资料，同时更新了过时的信息，以增强清晰度。书中还提供了Matlab代码示例，部分可在网络上下载，旨在使读者能够更直观地理解数据融合技术。数据融合是将来自多个传感器的数据整合，以提高信息的准确性和可靠性。本书适用于那些没有数据融合背景但对线性代数、微积分和简单概率有一定了解的电气工程师或计算机科学家。作者通过统一的统计框架，将神经网络、信号处理、统计估计、跟踪算法、计算机视觉和控制理论等多元领域的工具联系起来，揭示了不同方法之间的内在关系。书中强调了数据融合的概念性简洁，但同时也指出该领域在教育中面临的挑战。通过许多现实生活中的应用实例，如自动驾驶、环境监测、军事侦察等，读者可以深入理解数据融合在实际问题中的应用。此外，书中的例子和Matlab代码都以灰色背景突出显示，便于识别，并且标记了进阶材料，以帮助不同程度的读者按需学习。本书的主要内容包括： 1. 传感器及其校准：介绍传感器的工作原理、误差模型和校准方法，确保来自不同传感器的数据能够准确集成。 2. 数据融合架构：探讨各种数据融合架构，如层次融合、集中式融合和分布式融合，以及它们各自的优势和适用场景。 3. 常用算法：讲解统计估计、卡尔曼滤波、粒子滤波、贝叶斯估计等核心算法，以及如何在数据融合中应用这些算法进行信息处理。 4. 实际应用：通过案例研究，展示了数据融合在航空、交通、安全监控等领域的实践应用，使理论知识与实际操作相结合。 5. Matlab代码：提供可下载的Matlab代码，帮助读者实践和理解书中的算法，提高动手能力。 "Data Fusion- Concepts and Ideas" 是一个全面的学习资源，适合希望深入了解多传感器数据融合的学者和从业者。通过本书，读者不仅可以掌握数据融合的基本概念和技术，还能获得在实际项目中实施数据融合策略的宝贵经验。

1.3 Multi-sensor Data Fusion Strategies 3

Example 1.3. Multi-Modal Medical Imaging: Gain in Completeness. We con-

sider images obtained through Magnetic Resonance Imaging (MRI), Com-

puted Tomography (CT) and Positron Emission Tomography (PET). The

multi-sensor data fusion of all three images allows a surgeon to view “soft

tissue” information (MRI) in the context of “skeleton” or “bone information”

(CT) and in the context of “functional” or “physiological information” (PET).

1.3 Multi-sensor Data Fusion Strategies

As the above examples show, multi-sensor data fusion is a wide-ranging subject with

many different facets. In order to understand it better, and to become familiar with

its terminology, we shall consider it from three different points of view as suggested

by Boudjemaa and Forbes [5], Durrant-Whyte [7] and Dasarathy [6].

1.3.1 Fusion Type

Boudjemaa and Forbes [5] classify a multi-sensor data fusion system according to

what aspect of the system is fused:

Fusion across sensors. In this situation, a number of sensors nominally measure

the same property, as, for example, a number of temperature sensors measuring

the temperature of an object.

Fusion across attributes. In this situation, a number of sensors measure differ-

ent quantities associated with the same experimental situation, as, for example,

in the measurement of air temperature, pressure and humidity to determine air

refractive index.

Fusion across domains. In this situation, a number of sensors measure the same

attribute over a number of different ranges or domains. This arises, for example,

in the deﬁnition of a temperature scale.

Fusion across time. In this situation, current measurements are fused with his-

torical information, for example, from an earlier calibration. Often the current

information is not sufﬁcient to determine the system accurately and historical

information has to be incorporated to determine the system accurately.

Example 1.4. Flood Forecasting [23]. Water companies are under constant

pressure to reduce the frequency of combined sewer overﬂows to natural water

courses from urban drainage systems (UDS). The management of storm-water

through the UDS and similar applications require accurate real-time estimates

4 1 Introduction

of the rainfall. One way water companies have done this is to fuse measure-

ments of the rainfall made by ground-based rain gauges and a weather radar

system. This is “fusion across sensors” since, nominally, the two sensors mea-

sure the same property.

Example 1.5. Fire Detection [26] Urban-Rural Interface (URI) are zones

where forest and rural lands interface with homes, other buildings and in-

frastructures. In the URI, ﬁres are frequently due to the special nature of these

zones: a ﬁre can be the result of the action of a human on one side of the zone

or could arrive from the other side (wild land). In such cases, the process of

early detection and monitoring of a ﬁre event is important for efﬁcient control

of the ﬁre and if required, the evacuation of the area. In [26] a ﬁre is detected

using temperature, humidity and vision sensors. The temperature and humid-

ity sensors are deployed ”in-ﬁeld” in the URI zone while the vision sensor

is deployed ”out-ﬁeld” a distance from the URI zone. This is ”fusion-across-

attributes” since the sensors measure different quantities associated with the

same experimental situation.

1.3.2 Sensor Conﬁguration

Durrant-Whyte [7] classiﬁes a multi-sensor data fusion system according to its sen-

sor conﬁguration. There are three basic conﬁgurations:

Complementary. A sensor conﬁguration is called complementary if the sensors

do not directly depend on each other, but can be combined in order to give a more

complete image of the phenomenom under observation. Complementary sensors

help resolve the problem of incompleteness.

Competitive. A sensor conﬁguration is competitive if each sensor delivers an in-

dependent measurement of the same property. The aim of competitive fusion is

to reduce the effects of uncertain and erroneous measurements.

Cooperative. A cooperative sensor conﬁguration network uses the information

provided by two, or more, independent sensors to derive information that would

not be available from the single sensors.

The following example illustrates the complementary fusion of an infra-red and a

visible-light image.

6 1 Introduction

Example 1.9. Super-Resolution [22]. In super-resolution we cooperatively

fuse together two or more images of the same scene and taken by the same

sensor but from slightly different viewing angles. The result is a new image

which has higher spatial resolution than any of the input images.

1.3.3 Input/Output Characteristics

Dasarathy [6] classiﬁes a multi-sensor data fusion system according to its joint in-

put/output characteristics. Table 1.1 illustrates the types of inputs/outputs consid-

ered in this scheme. Note: Not all of the possible input-output combinations are rep-

resented in the Dasarathy scheme. Instead only combinations in which the output is

at the same semantic level, or one level higher than the input, are used.

Table 1.1 Dasarathy’s Input/Output Data Fusion Model.

Symbol Name Description/Example

DaI-DaO Data Input-Data Output Input data is smoothed/ﬁltered.

DaI-FeO Data Input-Feature Output Features are generated from the input data, e.g. edge

detectioninanimage.

FeI-FeO Feature Input-Feature Output Input features are reduced in number, or new features

are generated by fusing input features.

FeI-DeO Feature Input-Decision Out-

put

Input features are fused together to give output deci-

sion.

DeI-DeO Decision Input-Decision

Output

Multiple input decisions are fused together to give a

ﬁnal output decision.

It is sometimes useful to divide the DeI-DeO fusion model into two sub-models: a “soft” decision-

input model (denoted as DsI-DeO) in which each input decision is accompanied by a degree-of-

support, or reliability, value and a “hard” decision-input model (denoted as DhI-DeO) in which

the input decisions are not accompanied by any degree-of-support values.

Example 1.10. Multi-Modal Face and Palmprint Fusion [20]. Biometric

recognition systems often use face images or palmprint images. In general,

the recognition rates obtained using only face or palmprint images is relatively

low. However, signiﬁcantly better recognition rates are obtained by separately

extracting features from the face and palmprint images which are then fused

together. The fusion thus takes place in the FeI-FeO domain.

1.4 Formal Framework 7

1.4 Formal Framework

Multi-sensor data fusion systems are often of a size and a complexity that requires

the use of a formal framework [18] around which we may organize our knowledge.

One class of formal frameworks is the functional model. The functional model gives

an overview of the main processes, or functions, in the system. They are very im-

portant in designing, building and understanding large scale multi-sensor data fusion

systems. Table 1.2 lists some of the functional models which are used today [21]. For

systems which are smaller in size an architectural model which shows the individual

modules in the system and the ﬂow of data between modules is often more useful. In

this book we have opted for the architectual model of Luo-Kay [15] which consists

of a distributed network of autonomous modules, in which each module represents

a separate function in the data fusion system.

Table 1.2 Multi-Sensor Data Fusion Frameworks.

Name Description

Joint Defence

Laboratories (JDL)

The JDL model consists of several levels at which fusion can take place.

The levels are: Pre-Processing (Level 0); Object Reﬁnement (Level 1);

Situation Assessment (level 2); Threat Assessment (Level 3) and Pro-

cess Reﬁnement (Level 4).

Thompoulos The Thompoulos model consists of three modules, each module inte-

grates input data at a different level. The modules are: Signal Level

fusion, Evidence Level Fusion and Dynamics Level Fusion.

Pau The Pau model consists of a sequence of modules that must be com-

pleted before the ﬁnal fused output is established. The modules are:

Feature Extraction, Association Fusion, Sensor Attribute Fusion, Anal-

ysis, and Aggregation and Representation.

Waterfall The Waterfall model is similar to the Pau model except the system

is continuously updated with feedback information from the decision-

making module.

Omnibus The Omnibus model consists of four modules: Signal Processing mod-

ule, Pattern Recognition and Feature Extraction, Decision makng and

Context Processing and Control Resource Tasking. The modules oper-

ate in a closed loop.

Example 1.11. Physiological Measurements [3]. We consider two physiolog-

ical measurements made on a given patient: temperature and blood pressure.

The measurements are provided by two sensors: a thermometer and a ten-

siometer. The data sources are distributed and complementary. In forming a

diagnosis, the physician fuses together the two measurements. In the physi-

ological space of the patient the physician is said to perform “fusion across

attributes”.

剩余348页未读，继续阅读

xhhmelon

粉丝: 0
资源: 1

多传感器数据融合：概念与思想

concepts

数据融合：DataFusion现在已捐赠给Apache Arrow项目

DataFusion:在建 - 实时移动数据同步

Data Mining - Concepts and Techniques

Data Science-Concepts and Practice, 2nd edition

Data Mining - Concepts and Techniques (3rd Ed).pdf )

Data-Mining-Concepts-and-Techniques.rar_人工智能/神经网络/深度学习_Visual_C++_

data-engineering-concepts

Wideband Beamforming--Concepts and Techniques John Wiley

分布式网络存储Distributed File Systems--Concepts and Examples.PDF

最新资源