The thesis addresses the problem of detecting and identifying infrared LEDs, used as visual markers to support the subsequent estimation of relative pose. The problem was addressed through a multiclass object detection procedure, in which each LED must be localized within the image by means of a bounding box and associated with the correct identifying label. For this purpose, an experimental dataset was used, consisting of images acquired under different relative configurations between the camera and the target, containing up to nine LEDs arranged according to a known geometry. The YOLOv3 detector was trained in the MATLAB environment. The dataset was divided into training, validation and test sets, in order to evaluate the model’s ability to generalize to images not seen during the training phase. The performance was analyzed using quantitative metrics, supported by a qualitative verification of the results. The results obtained show that the use of neural networks represents a promising approach for automating the recognition of luminous markers.
La tesi affronta il problema del rilevamento e dell’identificazione di LED infrarossi, utilizzati come marker visivi per supportare la successiva stima della posa relativa. Il problema è stato trattato mediante una procedura di object detection multiclasse, in cui ciascun LED deve essere localizzato all’interno dell’immagine tramite una bounding box e associato alla corretta etichetta identificativa. A tale scopo è stato utilizzato un dataset sperimentale composto da immagini acquisite in diverse configurazioni relative tra camera e target, contenenti fino a nove LED disposti secondo una geometria nota. L’addestramento del detector YOLOv3 è stato svolto in ambiente MATLAB. Il dataset è stato suddiviso in training, validation e test set, così da valutare la capacità del modello di generalizzare su immagini non viste durante la fase di addestramento. Le prestazioni sono state analizzate mediante metriche quantitative affiancate da una verifica qualitativa dei risultati. I risultati ottenuti mostrano che l’impiego di reti neurali rappresenta un approccio promettente per automatizzare il riconoscimento dei marker luminosi.
Rilevamento e identificazione di marker luminosi mediante reti neurali per la stima della posa relativa in ambito aerospaziale
MANCUSO, CHIARA
2025/2026
Abstract
The thesis addresses the problem of detecting and identifying infrared LEDs, used as visual markers to support the subsequent estimation of relative pose. The problem was addressed through a multiclass object detection procedure, in which each LED must be localized within the image by means of a bounding box and associated with the correct identifying label. For this purpose, an experimental dataset was used, consisting of images acquired under different relative configurations between the camera and the target, containing up to nine LEDs arranged according to a known geometry. The YOLOv3 detector was trained in the MATLAB environment. The dataset was divided into training, validation and test sets, in order to evaluate the model’s ability to generalize to images not seen during the training phase. The performance was analyzed using quantitative metrics, supported by a qualitative verification of the results. The results obtained show that the use of neural networks represents a promising approach for automating the recognition of luminous markers.| File | Dimensione | Formato | |
|---|---|---|---|
|
Mancuso_Chiara.pdf
embargo fino al 12/10/2029
Dimensione
5.7 MB
Formato
Adobe PDF
|
5.7 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/20.500.12608/117276