Non-linear and non-Gaussian Sensor Fusion

Description of the special session:

Due to complex human motion patterns, uncertain sensors and complex environments, the problem of indoor positioning and navigation should be represented by a non-linear and non- Gaussian system in order to make accurate assumptions about the possible whereabouts of pedestrians. This special session will therefore focus on sensor fusion methods that explicitly try to solve the problem of indoor positioning and navigation in this context.

Novelty and motivation:

Many hybrid positioning and navigation approaches traditionally adopt a linear and Gaussian state space assumption, often utilizing classical Kalman filtering techniques or modeling sensor noise with Gaussian distributions. However, these assumptions can be limiting in various scenarios, particularly when dealing with more complex pedestrian motion patterns within 3D spatial models or when accounting for noise in RSSI-based distance estimation.

With the emergence of newer and more precise sensors, achieving an accurate representation of the state space becomes crucial. Nonlinear and non-Gaussian methods like particle filtering or fuzzy logic can offer robust fusion techniques that can better handle the complexities of real-world scenarios. Despite concerns about their perceived complexity, these methods should not be overlooked. In fact, their versatility presents significant opportunities, such as incorporating virtual sensors based on machine learning and AI into the fusion process.

In essence, while traditional linear and Gaussian assumptions have been prevalent, they may not suffice in capturing the intricacies of modern positioning and navigation challenges. Embracing nonlinear and non-Gaussian approaches opens up new avenues for harnessing the capabilities of advanced sensors and leveraging machine learning techniques to enhance fusion methods and improve overall accuracy.

Scope of the session:

This special session encompasses a broad range of topics related to advanced sensor fusion methods suited for non-linear and non-Gaussian environments. This includes, but is not limited to:

  • Probabilistic Methods:
    • Particle Filtering
    • Markov Chain Monte Carlo
    • Interacting Multiple Model Approaches
    • Kalman Filtering under Non-Gaussian Noise
  • Deep Fusion:
    • Reinforcement Learning
    • Representation Learning
    • Integration of Virtual Sensors and AI in Fusion Processes
  • Graph-Based Approaches:
    • Graph-Based Fusion Models
    • Graph Neural Networks
    • Dynamic Bayesian Networks
  • Theoretical and Mathematical Frameworks:
    • Information Fusion Theory
    • Possibility and Fuzzy Set Theory
    • Advanced Algorithms for Uncertain Sensor Data

This session aims to highlight the recent advancements and innovative approaches in sensor fusion that can effectively address the complexities encountered in dynamic and unpredictable environments. It will also explore the integration of novel technologies such as artificial intelligence and machine learning to enhance the capabilities and accuracy of fusion systems.


  • Sensor Fusion
  • Non-linear Dynamics
  • Non-Gaussian Noise
  • Particle Filtering
  • Multi-Modal Fusion
  • Graph-based Fusion
  • Deep Fusion

Session chairs:

  • Dr. Toni Fetzer (cronn GmbH)
  • Prof. Frank Deinzer (Technical University of Applied Sciences W├╝rzburg/Schweinfurt)