Shape Estimation: Detection and Measurement of Straps in 3D Point Clouds
Keywords
Loading...
Authors
Issue Date
2024-06-03
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
Baggage handling systems process baggage items at high speed. Therefore, automation is
needed to cope with the growing number of bags. The Bag Classifier module is built to extract
the properties of a baggage item using computer vision. This project studies the detection and
measurement of straps, that form a risk of getting entangled in the conveyor system. Annotation
and segmentation techniques on 2D and 3D data are studied. Three measurement algorithms
are proposed: sphere, box, and hybrid search. By running experiments on a synthetic and a
real-world dataset the performance of these search methods is evaluated. Results show that
sphere search is able to measure the length of 85% of the synthetic straps within the acceptable
error margin. Box search and hybrid search show the potential to achieve even higher results.
By closing the domain gap between synthetic and real-world straps, a productizable solution
might be found in the future. The outcomes of this project help the Bag Classifier module in
determining the risk of conveyability of a baggage item. Additionally, the methods studied and
developed for strap detection and measurement can be applied to other domains involving 3D
object detection and measurement.
Description
Citation
Supervisor
Faculty
Faculteit der Sociale Wetenschappen