This thesis presents the development of an automatic mechanical part recognition system based on neural networks. Image acquisition is performed using an Orbbec Femto Bolt RGB-D camera. The classification stage is implemented in MATLAB through a neural network trained on a dataset of images of mechanical components; the model is validated through final test trials to evaluate recognition performance under operational conditions. Once the part has been identified, the system leverages the depth image to obtain the object’s geometric dimensions. To improve usability and ensure consistency with a production environment, an operator panel was also developed in MATLAB, integrating acquisition, recognition, result visualization, and measurements, thus providing a clear interface for test analysis.
Questa tesi presenta lo sviluppo di un sistema di riconoscimento automatico di pezzi meccanici basato su reti neurali. L’acquisizione delle immagini viene effettuata tramite camera RGB-D Orbbec Femto Bolt. La fase di classificazione è realizzata in MATLAB mediante una rete neurale addestrata su un dataset di immagini dei componenti meccanici; il modello viene validato tramite prove finali di test per valutare la capacità di riconoscimento in condizioni operative. Una volta identificato il pezzo, il sistema sfrutta l’immagine in modalità depth per ricavare le dimensioni geometriche dell’oggetto. Per aumentare la fruibilità e la coerenza con l’ambiente produttivo, è stato inoltre sviluppato un pannello operatore, in ambiente matlab che integra acquisizione, riconoscimento, visualizzazione del risultato e misure, fornendo un’interfaccia chiara per l’analisi delle prove.
Applicazione di reti neurali convoluzionali a sistemi RGB-D per il riconoscimento e identificazione di componenti meccanici
REBUSTI, THOMAS
2025/2026
Abstract
This thesis presents the development of an automatic mechanical part recognition system based on neural networks. Image acquisition is performed using an Orbbec Femto Bolt RGB-D camera. The classification stage is implemented in MATLAB through a neural network trained on a dataset of images of mechanical components; the model is validated through final test trials to evaluate recognition performance under operational conditions. Once the part has been identified, the system leverages the depth image to obtain the object’s geometric dimensions. To improve usability and ensure consistency with a production environment, an operator panel was also developed in MATLAB, integrating acquisition, recognition, result visualization, and measurements, thus providing a clear interface for test analysis.| File | Dimensione | Formato | |
|---|---|---|---|
|
Rebusti_Thomas.pdf
accesso aperto
Dimensione
3.49 MB
Formato
Adobe PDF
|
3.49 MB | Adobe PDF | Visualizza/Apri |
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/20.500.12608/108063