In the field of Computer Vision the recognition of lines subjected to radial distortion plays a central role in many applications. Generally, the recognition of deformed lines is solved by estimating the correction of the radial distortion using a polynomial model. Although in the literature there are several models of distortion correction, in some applications it has been found that the proposed models are insufficient. It is therefore difficult to face the recognition of deformed lines in a generic way using distortion correction methods. In our experimentation we solve the problem of distorted lines detection through the analysis of preferences, using local clustering techniques in the preference space. Therefore, by exploiting Multi Model Fitting methods, we propose a general solution, which does not require any information on distortion, robust to noise and applicable to problems of non rigid structures detection. We present SoftLinkage, a first Non-Rigid Multi Model Fitting method capable of solving the problem of non rigid line detection through the analysis of preferences. The results of the experiments carried out confirm the effectiveness of SoftLinkage compared to methods that use distortion correction models, which verifies how these methods are limited in generic contexts where noise is present.
Nel campo della Computer Vision il riconoscimento di rette sottoposte a distorsione radiale ricopre un ruolo centrale in molteplici applicazioni. Classicamente il riconoscimento di rette deformate avviene stimando la correzione della distorsione radiale tramite un modello polinomiale. Nonostante in letteratura siano presenti diversi modelli di correzione della distorsione, in alcune applicazioni i modelli proposti risultano insufficienti. È quindi difficile affrontare in maniera generica il riconoscimento delle rette utilizzando un approccio che si basa sulla correzione della distorsione. Nella nostra sperimentazione risolviamo il problema del riconoscimento di rette distorte tramite metodi che sfruttano l'analisi delle preferenze, applicando tecniche di clustering locale sulle preferenze. Sfruttando quindi metodi di Multi Model Fitting proponiamo una soluzione generale, che non necessita di alcuna informazione sulla distorsione, robusta al rumore ed estendibile a problemi di riconoscimento di strutture geometriche distorte diverse da rette. Presentiamo il SoftLinkage, un primo approccio di metodo Multi Model Fitting Non Rigido capace di risolvere il problema del riconoscimento di rette distorte tramite l'analisi delle preferenze. I risultati degli esperimenti condotti confermano l'efficacia del SoftLinkage rispetto a metodi che usano modelli di correzione della distorsione, verificando come tali metodi si rivelano particolarmente limitati in contesti generici dove è presente il rumore.
Detezione di linee in immagini distorte tramite analisi delle preferenze
RUGGIANO, ENRICO
2021/2022
Abstract
In the field of Computer Vision the recognition of lines subjected to radial distortion plays a central role in many applications. Generally, the recognition of deformed lines is solved by estimating the correction of the radial distortion using a polynomial model. Although in the literature there are several models of distortion correction, in some applications it has been found that the proposed models are insufficient. It is therefore difficult to face the recognition of deformed lines in a generic way using distortion correction methods. In our experimentation we solve the problem of distorted lines detection through the analysis of preferences, using local clustering techniques in the preference space. Therefore, by exploiting Multi Model Fitting methods, we propose a general solution, which does not require any information on distortion, robust to noise and applicable to problems of non rigid structures detection. We present SoftLinkage, a first Non-Rigid Multi Model Fitting method capable of solving the problem of non rigid line detection through the analysis of preferences. The results of the experiments carried out confirm the effectiveness of SoftLinkage compared to methods that use distortion correction models, which verifies how these methods are limited in generic contexts where noise is present.File | Dimensione | Formato | |
---|---|---|---|
2022_04_Ruggiano.pdf
accessibile in internet per tutti
Dimensione
63.56 MB
Formato
Adobe PDF
|
63.56 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/10589/187678