Classificazione di oggetti in immagini attraverso il modello Bag of Visual Words

Generic object recognition is an important function of the human visual system and everybody finds it highly useful in their everyday life. For an artificial vision system it is a really hard, complex and challenging task because instances of the same object category can generate very different images, depending of different variables such as illumination conditions, the pose of an object, the viewpoint of the camera, partial occlusions, and unrelated background clutter. The purpose of this thesis is to develop a system that is able to classify objects in 2D images based on the context, and identify to which category the object belongs to. Given an image, the system can classify it and decide the correct categorie of the object. Furthermore the objective of this thesis is also to test the performance and the precision of different supervised Machine Learning algorithms in this specific task of object image categorization. Through different experiments the implemented application reveals good categorization performances despite the difficulty of the problem. However this project is open to future improvement; it is possible to implement new algorithms that has not been invented yet or using other techniques to extract features to make the system more reliable. This application can be installed inside an embedded system and after trained (performed outside the system), so it can become able to classify objects in a real-time. The information given from a 3D stereocamera, developed inside the department of Computer Engineering of the University of Bologna, can be used to improve the accuracy of the classification task. The idea is to segment a single object in a scene using the depth given from a stereocamera and in this way make the classification more accurate.

Analisi di algoritmi per reti di sensori radar uwb basati su misure sperimentali

In un sistema radar è fondamentale rilevare, riconoscere e cercare di seguire il percorso di un eventuale intruso presente in un’area di osservazione al fine ultimo della sicurezza, sia che si consideri l’ambito militare, che anche quello civile. A questo proposito sono stati fatti passi avanti notevoli nella creazione e sviluppo di sistemi di localizzazione passiva che possano rilevare un target (il quale ha come unica proprietà quella di riflettere un segnale inviato dal trasmettitore), in modo che esso sia nettamente distinto rispetto al caso di assenza dell’intruso stesso dall’area di sorveglianza. In particolare l’ultilizzo di Radar Multistatico (ossia un trasmettitore e più ricevitori) permette una maggior precisione nel controllo dell’area d’osservazione. Tra le migliori tecnologie a supporto di questa analisi vi è l’UWB (Ultra Wide-Band), che permette di sfruttare una banda molto grande con il riscontro di una precisione che può arrivare anche al centimetro per scenari in-door. L’UWB utilizza segnali ad impulso molto brevi, a banda larga e che quindi permettono una risoluzione elevata, tanto da consentire, in alcune applicazioni, di superare i muri, rimuovendo facilmente gli elementi presenti nell’ambiente, ossia il clutter. Quindi è fondamentale conoscere algoritmi che permettano la detection ed il tracking del percorso compiuto dal target nell’area. In particolare in questa tesi vengono elaborati nuovi algoritmi di Clustering del segnale ricevuto dalla riflessione sull’intruso, utilizzati al fine di migliorare la visualizzazione dello stesso in post-processing. Infine questi algoritmi sono stati anche implementati su misure sperimentali attuate tramite nodi PulsOn 410 Time Domain, al fine ultimo della rilevazione della presenza di un target nell’area di osservazione dei nodi.

Performance analysis of reconstruction algorithms for breast microwave imaging

The problem of localizing a scatterer, which represents a tumor, in a homogeneous circular domain, which represents a breast, is addressed. A breast imaging method based on microwaves is considered. The microwave imaging involves to several techniques for detecting, localizing and characterizing tumors in breast tissues. In all such methods an electromagnetic inverse scattering problem exists. For the scattering detection method, an algorithm based on a linear procedure solution, inspired by MUltiple SIgnal Classification algorithm (MUSIC) and Time Reversal method (TR), is implemented. The algorithm returns a reconstructed image of the investigation domain in which it is detected the scatterer position. This image is called pseudospectrum. A preliminary performance analysis of the algorithm vying the working frequency is performed: the resolution and the signal-to-noise ratio of the pseudospectra are improved if a multi-frequency approach is considered. The Geometrical Mean-MUSIC algorithm (GM- MUSIC) is proposed as multi-frequency method. The performance of the GMMUSIC is tested in different real life computer simulations. The performed analysis shows that the algorithm detects the scatterer until the electrical parameters of the breast are known. This is an evident limit, since, in a real life situation, the anatomy of the breast is unknown. An improvement in GM-MUSIC is proposed: the Eye-GMMUSIC algorithm. Eye-GMMUSIC algorithm needs no a priori information on the electrical parameters of the breast. It is an optimizing algorithm based on the pattern search algorithm: it searches the breast parameters which minimize the Signal-to-Clutter Mean Ratio (SCMR) in the signal. Finally, the GM-MUSIC and the Eye-GMMUSIC algorithms are tested on a microwave breast cancer detection system consisting of an dipole antenna, a Vector Network Analyzer and a novel breast phantom built at University of Bologna. The reconstruction of the experimental data confirm the GM-MUSIC ability to localize a scatterer in a homogeneous medium.