Related Concept Videos
Uncertainty: Confidence Intervals
Uncertainty: Overview
Propagation of Uncertainty from Random Error
Propagation of Uncertainty from Systematic Error
Modeling and Similitude
The Uncertainty Principle
You might also read
Related Articles
Articles linked to this work by shared authors, journal, and citation graph.
A Maximum Entropy Model of the Distribution of Dengue Serotype in Mexico.
Measuring the Uncertainty of Predictions in Deep Neural Networks with Variational Inference.
Design of a multiplex nested PCR for genotyping of the NSP4 from group A rotavirus.
Related Experiment Video
Updated: Nov 10, 2025

From Voxels to Knowledge: A Practical Guide to the Segmentation of Complex Electron Microscopy 3D-Data
Published on: August 13, 2014
From a Point Cloud to a Simulation Model-Bayesian Segmentation and Entropy Based Uncertainty Estimation for 3D
Christina Petschnigg1, Markus Spitzner1, Lucas Weitzendorf1
1BMW Group, Department of Factory Planning, Knorrstraße 147, 80788 Munich, Germany.
This study presents a new method for 3D factory modeling using Bayesian neural networks for point cloud segmentation. This approach significantly improves the accuracy of environment models for process simulations in industrial planning.
Area of Science:
- Industrial Engineering
- Computer Vision
- Robotics
Background:
- 3D modeling and process simulations are crucial for factory planning.
- Existing data in brownfield sites are often outdated and incomplete, hindering accurate model generation.
- Automated approaches for creating comprehensive factory models are lacking.
Purpose of the Study:
- To develop a methodical approach for automated 3D factory model generation from digitized indoor environments.
- To investigate the impact of uncertainty information from Bayesian segmentation on model accuracy.
- To improve the accuracy of environment models for simulation purposes.
Main Methods:
- Digitalization of large-scale indoor environments.
- Application of a Bayesian neural network for point cloud segmentation and object identification.
- Evaluation using a real-world dataset from an automotive production plant.
Main Results:
- The Bayesian segmentation network outperformed the frequentist baseline.
- Uncertainty information from the Bayesian framework enhanced environment model accuracy.
- Considerable increase in the accuracy of model placement within simulation scenes was achieved.
Conclusions:
- The proposed methodical approach enables automated generation of static environment or simulation models.
- Bayesian neural networks offer a superior solution for point cloud segmentation in industrial settings.
- The method significantly improves the accuracy and reliability of 3D factory models for planning and simulation.

