Hier finden Sie wissenschaftliche Publikationen aus den Fraunhofer-Instituten.

A first step towards automated image-based container inspections

: Kloever, Steffen; Kretschmann, Lutz; Jahn, Carlos

Kersten, W. ; TU Hamburg-Harburg; TU Hamburg-Harburg, Institut für Logistik und Unternehmensführung:
Data science and innovation in supply chain management : How data transforms the value chain
Berlin: epubli, 2020 (Proceedings of the Hamburg International Conference of Logistics (HICL) 29)
ISBN: 978-3-753123-46-2
Hamburg International Conference of Logistics (HICL) <2020, Online>
Fraunhofer IML ()
computer vision; deep learning; Container Transport; Visual Damage Recognition; logistics; industry 4.0; Digitalization; Innovation; supply chain management; artificial intelligence; Data Science

Purpose: The visual inspection of freight containers at depots is an essential part of the maintenance and repair process, which ensures that containers are in a suitable condition for loading and safe transport. Currently this process is done manually, which has certain disadvantages and insufficient availability of skilled inspectors can cause delays and poor predictability.
Methodology: This paper addresses the question whether instead computer vision algorithms can be used to automate damage recognition based on digital images. The main idea is to apply state-of-the-art deep learning methods for object recognition on a large dataset of annotated images captured during the inspection process in order to train a computer vision model and evaluate its performance.
Findings: The focus is on a first use case where an algorithm is trained to predict the view of a container shown on a given picture. Results show robust performance for this task.
Originality: The originality of this work arises from the fact that computer vision for damage recognition has not been attempted on a similar dataset of images captured in the context of freight container inspections.