【论文阅读】A Survey of Multisensor Fusion Techniques, Architectures and Methodologies

A Survey of Multisensor Fusion Techniques, Architectures and Methodologies

I. INTRODUCTION

Keywords-

1. multi-sensor fusion classification;
2. sensor topologies and configurations;
3. signal fusion;
4. track-to-track fusion;
5. symbol level fusion;
6. dempster-shafer evidential reasoning

Multi-sensor fusion combines data from different sensors into a common representation format.

Significant applications of multi-sensor fusion:

1. mobile robots [2,3,4,5],
2. defense systems (such as target tracking [2, 6, 7, 8]),
3. medicine [9, 10],
4. transportation systems [11,12]
5. industry [13,14,15].

II. MOTIVATION

Goal: achieve better operation of the system, also referred to as the synergistic effect.

Combining the data from a single sensor at different time intervals can also produce this effect.

Outcome: increased estimation accuracy and fault-tolerance.

III. SENSOR FUSION CATEGORIES

A. Complementary互补

Each sensor provides data about different aspects or attributes of the environment.

no dependency between the sensors.

B. Competitive竞争

Several sensors measure the same or similar attributes.

The measurements are taken independently, and can also include measurements at different time instants for a single sensor.

useful in fault tolerant architectures.

C. Co-operative合作

A common example is stereoscopic vision.

Dasarthy’s classification of multi-sensor fusion including:

• DAI-DAO Fusion => Data In Data Put
• DAI-FEO Fusion => Data In Feature Out
• FEI-FEO Fusion => Feature In Feature Out
• FEI-DEO FwiFusionon => Feature In Decision Out
• DEI-DEO FuFusionsiDn => Decision In Decision Out

IV. SENSOR FUSION TOPOLOGIES

A. Centralized Architecture

A single node handles the fusion process

B. Decentralized Architecture

Each of the sensor processes data at its node. Used in applications that are large and widespread.

C. Hierarchical Architecture

A combination of both centralized and distributed type.

Decentralized is better for: computational workload, communication bandwidth

Centralized is better for: higher accuracy

V. MULTI SENSOR FUSION MODELS

no specific model or architecture that is defmitive for all applications.

two most widely used architectures:

A. JDL Fusion Architecture

US Joint Directors of Laboratories.

five levels for data processing and a database. Components communicate through a bus interface.

Modules as below:

• Sources
• Man-Machine Interaction block
• Source Pre-Processing => Level 0
• Object Refinement => Level 1
• Situation Refmement => Level 2
• Threat Refinement => Level 3 （This level uses game theoretic techniques）
• Process Refmement => Level 4

Limitations:

data centric and hence hard to extend or reuse the applications based on this model. It is abstract and interpretation could be difficult.

B. Waterfall Fusion Process Model (WFFM)

deals with the low level processing of data.

acyclic (非循环) model. But modified WFFM provides for some feedback between the stages.

VI. SIGNAL LEVEL FUSION

Signal level fusion often has either or both of the following goals:

• Obtain a higher quality version of the input signals
• Obtain a feature or mid-level information about the system that a single measuring node cannot reveal.

commonly used signal fusion methodologies

A. Weighted Averaging

taking an average of the various sensor signals measuring a particular parameter of the environment

B. Kalman Filter

remove redundancy in the system and to predict the state of the system.

linear model.

current state of the system is dependent on the previous state.

two phases of state estimation with Kalman filter: 1) Predict phase 2) Update phase

C. Track to Track Fusion

has local tracks generated by distinct local sensors

D. Neural Networks

non-linear transfer functions and parallel processing capabilities. This can help in performing image fusion.

VII. DECISION LEVEL FUSTON

Same as Symbol level fusion.

combines several sub-decisions or features to yield a final or higher decision.

A. Dempster-Shafer Theory of evidence

D-S theory is a generalization of the probability theory.

B. Dempster-Shafer and Bayesian fusion comparison

Dempster-Shafer theory of evidence finds widespread use in human-robot interactive (HR!) applications. A review of a few applications of HRI can be found in [47].

Words

instrumental 仪器的

synergistic 协同的

spatial 空间的

temporal 世俗的

derive 得到

stereoscopic 立体的

corroborative 确凿的

concordant 匹配的

redundant 冗余

spacecraft 航天器

hierarchical 等级的

defmitive 决断的

kinematic 运动学

aircraft 飞机

missile 导弹

refinement 精炼

vulnerabilities 弱点

interpretation 解释

acyclic 非循环

reveal 揭示

heterogeneous 异质

undergo 经历

adaptive 适应的

discernment 辨别力

plausibility 似是而非

proposition 提议

terrain 地形

waterfall 瀑布