In recent years, the rapid technological development in the context of computer computing capacity coupled with the one in the field of artificial vision has made it possible to carry out accurate measurements deriving directly from digital image processing. In particular, on the market, there are several software able to provide a virtual reconstruction of a three-dimensional scene starting from the synchronous acquisition of a pair of images obtained from two stereoscopic cameras. Thanks to the high resolution and frequency of image acquisition guaranteed by today's technology, it is possible to measure objects in motion at high speeds and subject to high-frequency of vibrations. The basic idea of this work is the design of a prototype stereoscopic system of limited sizes capable of acquiring and processing images autonomously, managed by remote control. Before arriving at this point, a good part of the thesis is based on the comparison of some of the most widespread algorithms capable of performing the triangulation process in a pair of stereo images, in order to reach a state of the art in terms of uncertainties, complexity, timing of the various processing models. The intention is to provide a robust stereo vision method based on different algorithms able to compete with the Digital Image Correlation (DIC) technique, considered today as the guideline technique in stereovision analysis. Prior to this, a preliminary work was done on the theme of visual saliency. Salience is a developed visual skill of man that consists in identifying the only regions of interest within the visual field in order to skim the amount of data that the brain has to process. In the same way, in the field of artificial vision, the idea is to eclipse the parts of the images considered irrelevant and then proceed with the processing of the only salient parts. The expectation is to encounter a reduction in computational complexity and an increase in measurement accuracy. After obtaining a state of the art of the different triangulation methods, a small-sized stereo system is devised, projected to be mounted on an unmanned aerial vehicle (UAV) for remote measurements. Finally, once the new stereo system has been completed, a sequence of acquisitions is performed to test its efficiency in static and dynamic measurements on the ground to qualify it with respect to the state of the art previously obtained. Likewise, in the case of dynamic measurements of vibratory motions, an analysis technique alternative to Digital Image Correlation is developed with the stereoscopic triangulation algorithms. The results obtained after the various tests are to be considered positive and therefore allow to characterize metrologically the designed stereovision system.
Negli ultimi anni, l’incremento esponenziale della capacità di calcolo dei computer affiancato allo sviluppo tecnologico nel campo della visione artificiale ha reso possibile l’idea di effettuare misurazioni accurate derivanti direttamente dalla rielaborazione di immagini digitali. In particolare, esistono diversi software in grado di ricostruire una scena tridimensionale partendo dalla acquisizione sincrona di una coppia di frames ottenuti da due telecamere stereoscopiche. Grazie all’elevata risoluzione e frequenza di acquisizione delle immagini garantite dalla tecnologia di oggi, è possibile effettuare misurazioni di oggetti in moto ad elevate velocità e soggetti a vibrazioni ad alta frequenza. L’idea di fondo di questo lavoro è la progettazione di un prototipo sistema stereoscopico di dimensioni compatte capace di acquisire e elaborare immagini autonomamente, gestito tramite controllo remoto. Prima di arrivare a questo punto, una buona parte della tesi è basata sul confronto di alcuni tra i più diffusi algoritmi capaci di effettuare il processo di triangolazione in una coppia di immagini stereo, in modo da raggiungere uno stato dell’arte in termine di incertezze, complessità, tempistiche dei vari modelli di processing. L’intenzione è quella di fornire un robusto metodo di visione stereo capace di competere con la Digital Image Correlation (DIC), considerata ad oggi come la tecnica di riferimento per analisi di questo tipo. Precedentemente a questo, è stato fatto un lavoro preliminare sul tema della visual saliency. La salienza è un’abilità visiva sviluppata dell’uomo che consiste nell'identificare solo le regioni di interesse all'interno del campo visivo col fine di scremare la quantità di dati che il cervello deve rielaborare. Allo stesso modo nell'ambito della visione artificiale, l’idea è quella di eclissare le parti delle immagini ritenute non rilevanti per poi procedere con l’elaborazione delle sole parti salienti. L’aspettativa è quella di riscontrare una riduzione della complessità computazionale e un incremento della precisione della misura. Dopo aver ottenuto uno stato dell’arte dei diversi metodi di triangolazione, viene ideato un sistema stereo di piccole dimensioni progettato nell'ottica di essere montato su di un aeromobile a pilotaggio remoto per effettuare misurazioni a distanza. Infine, ultimato il nuovo sistema stereo compatto, viene eseguita una sequenza di acquisizioni per testarne l’efficienza in misurazioni statiche e dinamiche a terra, in modo da qualificarlo rispetto allo stato dell’arte precedentemente ottenuto. Anche nel caso di misurazioni dinamiche di moti vibratori, viene ideata una tecnica di analisi con gli algoritmi stereoscopici di triangolazione alternativa alla Digital Image Correlation. I risultati ottenuti dopo i vari test sono di carattere positivo e consento quindi di caratterizzare metrologicamente il sistema stereo progettato.
Performance and uncertainty analysis of stereoscopy algorithms for the development of a portable stereovision rig
VALLI, FABIO BRIAN
2017/2018
Abstract
In recent years, the rapid technological development in the context of computer computing capacity coupled with the one in the field of artificial vision has made it possible to carry out accurate measurements deriving directly from digital image processing. In particular, on the market, there are several software able to provide a virtual reconstruction of a three-dimensional scene starting from the synchronous acquisition of a pair of images obtained from two stereoscopic cameras. Thanks to the high resolution and frequency of image acquisition guaranteed by today's technology, it is possible to measure objects in motion at high speeds and subject to high-frequency of vibrations. The basic idea of this work is the design of a prototype stereoscopic system of limited sizes capable of acquiring and processing images autonomously, managed by remote control. Before arriving at this point, a good part of the thesis is based on the comparison of some of the most widespread algorithms capable of performing the triangulation process in a pair of stereo images, in order to reach a state of the art in terms of uncertainties, complexity, timing of the various processing models. The intention is to provide a robust stereo vision method based on different algorithms able to compete with the Digital Image Correlation (DIC) technique, considered today as the guideline technique in stereovision analysis. Prior to this, a preliminary work was done on the theme of visual saliency. Salience is a developed visual skill of man that consists in identifying the only regions of interest within the visual field in order to skim the amount of data that the brain has to process. In the same way, in the field of artificial vision, the idea is to eclipse the parts of the images considered irrelevant and then proceed with the processing of the only salient parts. The expectation is to encounter a reduction in computational complexity and an increase in measurement accuracy. After obtaining a state of the art of the different triangulation methods, a small-sized stereo system is devised, projected to be mounted on an unmanned aerial vehicle (UAV) for remote measurements. Finally, once the new stereo system has been completed, a sequence of acquisitions is performed to test its efficiency in static and dynamic measurements on the ground to qualify it with respect to the state of the art previously obtained. Likewise, in the case of dynamic measurements of vibratory motions, an analysis technique alternative to Digital Image Correlation is developed with the stereoscopic triangulation algorithms. The results obtained after the various tests are to be considered positive and therefore allow to characterize metrologically the designed stereovision system.File | Dimensione | Formato | |
---|---|---|---|
2018_12_Valli.pdf
solo utenti autorizzati dal 28/11/2019
Descrizione: Testo della tesi
Dimensione
62.76 MB
Formato
Adobe PDF
|
62.76 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/143862