In this thesis, we have studied the possibility of using eye data about pupil size and gaze position in the context of the so-called “soft biometrics” (which does not require extremely high success rates). The two classic biometric processes have been considered, namely identification (i.e., recognizing an individual within a set) and verification (i.e., confirming or denying the identity declared by an individual). In experiments carried out in three sessions, some subjects were asked to freely look at an animation in which three figures (a circle, a square and a triangle) move according to random paths. Data were acquired using the Eye Tribe eye tracker with a 30 Hz sampling frequency, and specific software was implemented in C# to create the animation and capture eye data. Two different animations were realised: a mono-colour version, in which all three figures are red, and a multicolour version, in which the three shapes are different colours. In addition, the animation was subdivided into three phases: in the first only the circle is present, in the second there are the circle and the square, and in the third all three shapes are displayed). The analysis was performed considering the three phases separately, two consecutive phases and all three phases together. In the context of Machine Learning, four classifiers (Naive Bayes, Neural Network, Classification Tree and Random Forest) and two different sampling methods have been used. The obtained results show that there is no substantial difference between the two versions (mono and multicolour), although the mono-colour case provides slightly better outcomes, and that the second and the third phases considered together are the best case for what concerns animation phases. Overall, results are encouraging and suggest that eye features, with the considered stimuli, can provide useful information for both identification and verification.
Biometria basata su tracciamento oculare: uno studio sull'osservazione libera di elementi in movimento. In questa tesi abbiamo studiato la possibilità di utilizzare dati oculari relativi a dimensione delle pupille e posizione dello sguardo nel campo della cosiddetta “biometria soft” (la quale non richiede percentuali di successo estremamente elevate). Sono stati considerati i due classici processi biometrici, ossia l’identificazione (riconoscere un individuo all’interno di un insieme) e la verifica (confermare o negare l’identità dichiarata da un individuo). In esperimenti effettuati in tre diverse sessioni, è stato chiesto ad alcuni soggetti di guardare liberamente un’animazione in cui tre figure (un cerchio, un quadrato ed un triangolo) appaiono in sequenza e si muovono secondo percorsi casuali. I dati sono stati acquisiti utilizzando l’eye tracker Eye Tribe con una frequenza di campionamento di 30 Hz. Per l’implementazione dell’animazione e la registrazione dei dati oculari è stato sviluppato un apposito software in C#. L’animazione è stata realizzata in due versioni: una monocolore, in cui tutte e tre le figure sono di colore rosso, ed una multicolore, in cui le tre figure hanno colori differenti. Inoltre l’animazione è stata suddivisa in tre fasi: nella prima è presente solo il cerchio, nella seconda ci sono cerchio e quadrato e nella terza sono visualizzate tutte e tre le figure. L’analisi è stata effettuata considerando le tre fasi separatamente, due fasi consecutive e tutte e tre le fasi insieme. Nel contesto del Machine Learning, sono stati utilizzati quattro classificatori (Naive Bayes, Neural Network, Classification Tree e Random Forest) e due metodi di campionamento. I risultati ottenuti mostrano che tra le due versioni, mono e multicolore, non c’è sostanziale differenza, anche se si ottengono risultati leggermente migliori con la versione monocolore, mentre l’analisi delle fasi dell’animazione evidenzia risultati migliori considerando la seconda e la terza fase. Nel complesso, i risultati sono incoraggianti e suggeriscono che le caratteristiche dell’occhio, con gli stimoli considerati, possono fornire informazioni utili sia per il processo di identificazione che per quello di verifica.
Gaze-Based Biometrics: a Study on Free Observation of Moving Targets
AZZOLINA, STEFANO
2016/2017
Abstract
In this thesis, we have studied the possibility of using eye data about pupil size and gaze position in the context of the so-called “soft biometrics” (which does not require extremely high success rates). The two classic biometric processes have been considered, namely identification (i.e., recognizing an individual within a set) and verification (i.e., confirming or denying the identity declared by an individual). In experiments carried out in three sessions, some subjects were asked to freely look at an animation in which three figures (a circle, a square and a triangle) move according to random paths. Data were acquired using the Eye Tribe eye tracker with a 30 Hz sampling frequency, and specific software was implemented in C# to create the animation and capture eye data. Two different animations were realised: a mono-colour version, in which all three figures are red, and a multicolour version, in which the three shapes are different colours. In addition, the animation was subdivided into three phases: in the first only the circle is present, in the second there are the circle and the square, and in the third all three shapes are displayed). The analysis was performed considering the three phases separately, two consecutive phases and all three phases together. In the context of Machine Learning, four classifiers (Naive Bayes, Neural Network, Classification Tree and Random Forest) and two different sampling methods have been used. The obtained results show that there is no substantial difference between the two versions (mono and multicolour), although the mono-colour case provides slightly better outcomes, and that the second and the third phases considered together are the best case for what concerns animation phases. Overall, results are encouraging and suggest that eye features, with the considered stimuli, can provide useful information for both identification and verification.È consentito all'utente scaricare e condividere i documenti disponibili a testo pieno in UNITESI UNIPV nel rispetto della licenza Creative Commons del tipo CC BY NC ND.
Per maggiori informazioni e per verifiche sull'eventuale disponibilità del file scrivere a: unitesi@unipv.it.
https://hdl.handle.net/20.500.14239/25467