This master thesis delves into Human Pose Estimation (HPE) and Depth Estimation (DE), two crucial tasks in Computer Vision applications. Human Pose Estimation aims to identify and localize keypoints of the human body within an image, while Depth Estimation focuses on predicting the distance of objects from a camera without using additional sensors. In today's industrial landscape, the integration of edge computing plays an important role in advancing industry progress, potentially accelerating the transition to Industry 5.0. Leveraging on dedicated cameras made by Luxonis, with on-chip machine learning, as edge devices, this research, conducted in collaboration with beanTech, an IT company, aims to explore novel and cost-effective solutions to enhance quality control and safety in industrial settings. This work aims to evaluate various Deep Learning models for their applicability in achieving these objectives, highlighting their pros and cons, and to develop systems that can effectively serve industrial needs. In particular, this experimental thesis focuses to develop a system capable of detecting when workers are in hazardous positions or have fallen, and issuing an alert accordingly. Furthermore, it seeks to test Depth Estimation models to estimate the quantity of materials on a conveyor belts, thereby optimizing industrial operations.
Questa tesi magistrale approfondisce le tecniche di Human Pose Estimation (HPE) and Depth Estimation (DE), due aspetti cruciali nelle applicazioni di Computer Vision. Gli algoritmi di Human Pose Estimation mirano a identificare e localizzare i punti chiave del corpo umano all'interno di un'immagine, mentre quelli di Depth Estimation si concentrano sulla stima della distanza degli oggetti da una telecamera, senza l'uso di sensori aggiuntivi. Nel panorama industriale odierno, l'integrazione di soluzioni edge computing gioca un ruolo importante nell'avanzamento del progresso industriale, potenzialmente accelerando la transizione verso l'Industria 5.0. Sfruttando delle telecamere dedicate prodotte da Luxonis, che permettono di eseguire modelli di machine learning direttamente sul processore integrato, come dispositivi a basso costo, questa ricerca, condotta in collaborazione con beanTech, un'azienda IT, mira a esplorare soluzioni innovative e economiche per migliorare il controllo di qualità e la sicurezza negli ambienti industriali. Questo lavoro mira ad esaminare diversi modelli di Deep Learning al fine di valutarne l'applicabilità nel raggiungimento di tali obiettivi, evidenziandone vantaggi e svantaggi, e a sviluppare sistemi che possano servire efficacemente le esigenze industriali. In particolare, questa tesi sperimentale si concentra nello sviluppare un sistema capace di rilevare quando gli operai si trovano in posizioni pericolose o sono caduti, e di allertare di conseguenza. Inoltre, essa mira a testare modelli di Depth Estimation per stimare la quantità di materiali su nastri trasportatori, ottimizzando così il processo produttivo industriale.
Human Pose and Depth Estimation on Affordable VPU-based Platforms for Enhancing Industrial Safety and Process Control
GREGORI, SIMONE
2023/2024
Abstract
This master thesis delves into Human Pose Estimation (HPE) and Depth Estimation (DE), two crucial tasks in Computer Vision applications. Human Pose Estimation aims to identify and localize keypoints of the human body within an image, while Depth Estimation focuses on predicting the distance of objects from a camera without using additional sensors. In today's industrial landscape, the integration of edge computing plays an important role in advancing industry progress, potentially accelerating the transition to Industry 5.0. Leveraging on dedicated cameras made by Luxonis, with on-chip machine learning, as edge devices, this research, conducted in collaboration with beanTech, an IT company, aims to explore novel and cost-effective solutions to enhance quality control and safety in industrial settings. This work aims to evaluate various Deep Learning models for their applicability in achieving these objectives, highlighting their pros and cons, and to develop systems that can effectively serve industrial needs. In particular, this experimental thesis focuses to develop a system capable of detecting when workers are in hazardous positions or have fallen, and issuing an alert accordingly. Furthermore, it seeks to test Depth Estimation models to estimate the quantity of materials on a conveyor belts, thereby optimizing industrial operations.File | Dimensione | Formato | |
---|---|---|---|
Gregori_Simone.pdf
embargo fino al 10/07/2025
Dimensione
18 MB
Formato
Adobe PDF
|
18 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/20.500.12608/66606