A Fast Deep Learning Based Approach for Unsupervised Anomaly Detection in 3D Data
Autores: Sergio Presa Feijoo
Fecha: 28.10.2022
Abstract
Anomaly detection is an important method in industrial manufacturing environments for defect detection, and consequently, for the quality requirements attainment. There is a strong trend to automate inspection systems using Artificial Intelligence techniques, especially through the use of 2D information. However, for dimensional inspection, it is necessary to have 3D data that provide detailed geometric information of the object. In this paper, we present a novel Deep Learning based approach for fast anomaly detection applicable to industrial domains by using point cloud voxelization. We evaluate our approach on a custom dataset in which our model achieves high accuracy, comparable to the state-of-the-art 3D Deep Learning models, while being faster than traditional anomaly detection methods.
BIB_text
title = {A Fast Deep Learning Based Approach for Unsupervised Anomaly Detection in 3D Data},
pages = {6-13},
keywds = {
anomaly detection, artificial neural networks, geometry processing, intelligent manufacturing systems, unsupervised learning, Deep learning, Point cloud compression
}
abstract = {
Anomaly detection is an important method in industrial manufacturing environments for defect detection, and consequently, for the quality requirements attainment. There is a strong trend to automate inspection systems using Artificial Intelligence techniques, especially through the use of 2D information. However, for dimensional inspection, it is necessary to have 3D data that provide detailed geometric information of the object. In this paper, we present a novel Deep Learning based approach for fast anomaly detection applicable to industrial domains by using point cloud voxelization. We evaluate our approach on a custom dataset in which our model achieves high accuracy, comparable to the state-of-the-art 3D Deep Learning models, while being faster than traditional anomaly detection methods.
}
isbn = {978-166548158-8},
date = {2022-10-28},
}
