Monte Carlo Approaches

For more info about our services contact : help@bestpfe.com

Table of contents

Chapter 1 Introduction
1.1 Examples and Motivations
1.2 Problematic
1.3 Literature snapshot and current limitations
1.4 Contributions
1.5 Document Structure
Part I Bayesian Filtering
Chapter 2 The Filtering Problem
2.1 Partially Observed Dynamical System: The Filtering Problem
2.2 Bayesian Filter
2.3 Kalman Filter
2.4 Other Approaches
2.4.1 Analytical Approaches
2.4.2 Monte Carlo Approaches
2.5 Conclusion
Chapter 3 Particle Filtering
3.1 Monte Carlo methods
3.2 Importance Sampling Principle
3.3 Sequential Importance Sampling
3.4 Particle Filter
3.5 Regularized Particle Filter
3.6 Conclusion
Part II Single-Target Behavioral Tracking
Chapter 4 Algorithms for estimating pedestrian’s behavior
4.1 Introduction
4.2 Motion Tracking
4.2.1 Simple motion models
4.2.2 Motion models handling static obstacles
4.2.3 Motion models handling dynamic obstacles
4.2.3.1 Cellular-based motion models
4.2.3.2 Physics-based motion models
4.2.4 Overview of pedestrian tracking related works
4.3 Activity Recognition
4.3.1 Overview
4.3.1.1 Low-level activities
4.3.1.2 High-level activities
4.3.2 Graphical model based approaches
4.3.3 Syntactic approaches
4.3.4 Description-based approaches
4.4 Simultaneous Tracking and Activity Recognition
4.5 Conclusion
Chapter 5 Inferring pedestrian behaviors from agent-based simulations
5.1 Introduction
5.1.1 The STAR approach and current limitations
5.1.2 Agent-based behavioral simulators: an alternative to graphical models
5.1.3 Contributions
5.2 Autonomous agent-based behavioral simulation
5.2.1 Environment’s Service Representation
5.2.2 Action Selection Mechanisms
5.2.3 High-level Planners
5.2.4 Summary
5.3 Agent-Based Behavior Tracking
5.3.1 Process Overview
5.3.2 System Dynamics
5.3.3 Observation Model
5.4 Implementation
5.4.1 Simulator
5.4.2 Interactions with objects
5.4.3 System Architecture
5.5 Experimental Evaluations
5.5.1 Performance metrics
5.5.2 Virtual-world based experiments
5.5.2.1 Experimental Setup
5.5.2.2 Scenario 1: target with a xed motivation
5.5.2.3 Scenario 2: target with a varying motivation
5.5.2.4 Scenario 3: exogenous events
5.5.3 Real-world based experiments
5.5.3.1 Experimental Setup
5.5.3.2 Scenario 1
5.5.3.3 Scenario 2
5.6 Conclusion and Discussion
5.6.1 Contributions
5.6.2 Research directions
Part III Multi-Target Behavioral Tracking
Chapter 6 The Multi-Target Tracking Problem
6.1 Introduction
6.2 Problem Formulation
6.2.1 System Dynamics
6.2.2 Observation Model
6.2.2.1 Observation-Generation-related Assumptions
6.2.2.2 Observation Generation Procedure
6.2.3 Summary
6.3 Data-Association Problem
6.3.1 Global Nearest Neighbor
6.3.2 Multiple Hypothesis Tracker
6.3.2.1 Gating Procedure
6.3.2.2 Generation of Hypotheses
6.3.2.3 Hypothesis Probability Computation
6.3.2.4 Hypothesis Reduction Techniques
6.3.3 Joint Probabilistic Data Association Filter
6.3.3.1 Feasible (Joint Association) Hypothesis Generation
6.3.3.2 Computation of the jk coecients
6.4 Management of Targets’ Interactions
6.4.1 Observation-based Interactions
6.4.1.1 Approaches with observation-based potential functions
6.4.1.2 Learning-based Approaches
6.4.1.3 Interaction model-based Approaches
6.4.2 Dynamics-based Interactions
6.4.2.1 Approaches with state-based potential functions
6.4.2.2 Interaction Model-based Approaches
6.4.3 Summary
6.5 Conclusion
Chapter 7 Tracking Multiple Interacting Targets – An Interaction-Model Based Factored
7.1 Introduction
7.2 Problem Statement
7.3 Dynamics-based Interaction Representation
7.4 Estimation of Targets’ Predicted Distributions
7.4.1 Bayesian formulation of target predicted distributions
7.4.2 Exploiting the locality of interactions
7.4.3 Probability distribution aggregation
7.4.3.1 Overview
7.4.3.2 Eect-based partitioning
7.4.3.3 Estimation of the predicted distribution
7.4.4 Heuristics for aggregating probability distributions
7.4.4.1 Denition of the anity function
7.4.4.2 Partitioning of the probability distribution
7.4.4.3 Computing the representative states
7.4.5 Summary
7.5 Implementation Using Particle Filtering
7.5.1 Monte Carlo JPDAF
7.5.1.1 Soft-gating procedure
7.5.1.2 MC-JPDAF algorithm
7.5.1.3 Summary
7.5.2 Dealing with non-covered areas in (MC-)JPDAF
7.5.3 Tracking interacting targets with JPDA-like lter
7.5.3.1 Managing targets’ interactions
7.5.3.2 Particle reduction policy
7.5.3.3 Summary
7.5.4 Multiple-representative-based soft-gating
7.5.5 Summary
7.5.6 Complexity analysis
7.6 Experimental Evaluations
7.6.1 Simulator
7.6.2 Performance metrics
7.6.3 Single-representative-based evaluations
7.6.3.1 Experimental setup
7.6.3.2 Impact analysis of
7.6.3.3 Impact analysis of N
7.6.3.4 Impact analysis of K
7.6.3.5 Summary
7.6.4 Multiple-representative-based evaluations: an illustrative scenario
7.6.4.1 Experimental setup
7.6.4.2 Impact of multiple representatives at the prediction step .
7.6.4.3 Impact of multiple representatives at correction step
7.6.4.4 Summary
7.6.5 Multiple-representative-based evaluations: a complex scenario
7.6.5.1 Experimental setup
7.6.5.2 Single versus multiple representative(s)
7.6.5.3 Impact analysis of the clustering methodology
7.6.5.4 Impact analysis of the reduction policy
7.6.5.5 Impact analysis of N
7.6.5.6 Impact analysis of K
7.6.5.7 Coupling with an external re-identication module
7.7 Conclusion and Discussion
Chapter 8 Conclusion
8.1 Contributions
8.1.1 Simultaneous tracking and activity recognition
8.1.2 Management of target interactions
8.2 Future Work
8.2.1 Real-world evaluations
8.2.2 Target re-identication
8.2.3 Sensor control decision policies
8.2.4 Beyond behavioral tracking
8.3 Application domains
8.3.1 Behavioral model calibration
8.3.2 Telemedicine monitoring service
Bibliography