Luggage management is a core part of the ground handling operations performed by airports which requires meticulous care. A Unit Load Device (ULD) is a container which allows a more efficient load and handle of cargo for aircrafts. Their control and supervision is a critical challenge for airlines and ground service providers. The aim of this work is to develop a pipeline for the detection, recog- nition and identification of ULD IDs, which purpose is, along with other projects, to improve the airfield operations regarding the management of luggage transportation in the Frankfurt airport by leveraging the use of AI and Computer Vision systems. In particular, the pipeline is composed by a Character Region Awareness model for Text Detection (CRAFT) to detect and extract text from images of the environment, a four-stage Scene Text Recognition (STR) model to rec- ognize the text detected in the images extracted in the previous step and a filter based on a regular expression to identify the ULD IDs in the recognized text. Moreover, a dataset of 700 images of ULD IDs was constructed with the purpose of fine tuning different versions of the STR model. Different data augmentation techniques were also consequently applied to a training set of 400 images to face the issue of data scarcity. Following the proposed approach, the model achieved an accuracy of 90%.
Detection, recognition and identification of Unit Load Devices for aircrafts
PATARNELLO, LUCA
2022/2023
Abstract
Luggage management is a core part of the ground handling operations performed by airports which requires meticulous care. A Unit Load Device (ULD) is a container which allows a more efficient load and handle of cargo for aircrafts. Their control and supervision is a critical challenge for airlines and ground service providers. The aim of this work is to develop a pipeline for the detection, recog- nition and identification of ULD IDs, which purpose is, along with other projects, to improve the airfield operations regarding the management of luggage transportation in the Frankfurt airport by leveraging the use of AI and Computer Vision systems. In particular, the pipeline is composed by a Character Region Awareness model for Text Detection (CRAFT) to detect and extract text from images of the environment, a four-stage Scene Text Recognition (STR) model to rec- ognize the text detected in the images extracted in the previous step and a filter based on a regular expression to identify the ULD IDs in the recognized text. Moreover, a dataset of 700 images of ULD IDs was constructed with the purpose of fine tuning different versions of the STR model. Different data augmentation techniques were also consequently applied to a training set of 400 images to face the issue of data scarcity. Following the proposed approach, the model achieved an accuracy of 90%.File | Dimensione | Formato | |
---|---|---|---|
Patarnello_Luca.pdf
accesso riservato
Dimensione
5.05 MB
Formato
Adobe PDF
|
5.05 MB | Adobe PDF |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/43321