6 Data-driven state prediction and target tracking
Abstract
In Bayesian filtering, the time-varying state of a target is estimated using recursive prediction-update cycles, given a stream of measurements that are received through a sensor. Each of these steps involves certain modelling assumptions that need to reflect the physical properties of the target and the sensing system. Most importantly, the target is assumed to evolve according to a specific dynamical model that might describe changes (or the lack thereof) in position, heading, temperature, or other properties of interest. Each modelling choice can imply very specific restrictions on the physics of the target. A grave model mismatch might lead to high estimation errors or even track loss. In this chapter, neural networks are used for the state prediction, hence yielding an alternative paradigm to standard analytical modelling. It is further discussed which use cases exist for both analytic and data-driven modelling and how neural networks can be successfully included in the Bayes recursion.
Abstract
In Bayesian filtering, the time-varying state of a target is estimated using recursive prediction-update cycles, given a stream of measurements that are received through a sensor. Each of these steps involves certain modelling assumptions that need to reflect the physical properties of the target and the sensing system. Most importantly, the target is assumed to evolve according to a specific dynamical model that might describe changes (or the lack thereof) in position, heading, temperature, or other properties of interest. Each modelling choice can imply very specific restrictions on the physics of the target. A grave model mismatch might lead to high estimation errors or even track loss. In this chapter, neural networks are used for the state prediction, hence yielding an alternative paradigm to standard analytical modelling. It is further discussed which use cases exist for both analytic and data-driven modelling and how neural networks can be successfully included in the Bayes recursion.
Chapters in this book
- Frontmatter I
- Dedication V
- Foreword VII
- Preface IX
- Contents XXIII
-
Part I Artificially intelligent sensing
- 1 Sound source classification using deep learning image classification networks 3
- 2 Advances in array calibration 19
- 3 Optimal sensor placement using genetic algorithms 37
-
Part II Data-driven learning algorithms
- 4 Machine learning for electronic intelligence 59
- 5 Trajectory optimization with reinforcement learning 75
- 6 Data-driven state prediction and target tracking 93
-
Part III Discussion of advanced applications
- 7 Track-before-detect for passive radar 109
- 8 Data fusion for reconnaissance of radio nuclides 125
-
Part IV Managing multifunctional sensors
- 9 Multifunction RF sensor management 147
- 10 Perspectives on artificial intelligence in sensor resources management 161
- 11 Intelligent sensor network management 175
-
Part V Quantum algorithms for data fusion
- 12 Quantum algorithms for data fusion 193
- 13 Indistinguishability and anti-symmetry in multiple target tracking 205
- 14 Quantum computing for solving data association problems 225
-
Part VI Issues of certification and ethical alignment
- 15 Sensor data integrity in maritime multi-sensor networks 247
- 16 Explainable and certifiable AI 265
- 17 Ethical issues of AI-based sensing 275
- List of abbreviations
- Index 293
Chapters in this book
- Frontmatter I
- Dedication V
- Foreword VII
- Preface IX
- Contents XXIII
-
Part I Artificially intelligent sensing
- 1 Sound source classification using deep learning image classification networks 3
- 2 Advances in array calibration 19
- 3 Optimal sensor placement using genetic algorithms 37
-
Part II Data-driven learning algorithms
- 4 Machine learning for electronic intelligence 59
- 5 Trajectory optimization with reinforcement learning 75
- 6 Data-driven state prediction and target tracking 93
-
Part III Discussion of advanced applications
- 7 Track-before-detect for passive radar 109
- 8 Data fusion for reconnaissance of radio nuclides 125
-
Part IV Managing multifunctional sensors
- 9 Multifunction RF sensor management 147
- 10 Perspectives on artificial intelligence in sensor resources management 161
- 11 Intelligent sensor network management 175
-
Part V Quantum algorithms for data fusion
- 12 Quantum algorithms for data fusion 193
- 13 Indistinguishability and anti-symmetry in multiple target tracking 205
- 14 Quantum computing for solving data association problems 225
-
Part VI Issues of certification and ethical alignment
- 15 Sensor data integrity in maritime multi-sensor networks 247
- 16 Explainable and certifiable AI 265
- 17 Ethical issues of AI-based sensing 275
- List of abbreviations
- Index 293
