Predicting Visual Search Targets via Eye-Tracking Data

The purpose of this work is to train a binary support vector machine to develop a learning mechanism to predict visual search targets and eventually over an unknown set of targets. We used a dataset that contains fixation points of 18 participants searching for a particular image within three types of collages. After plotting the visual patterns of each of the users, we compare the features of each of the images the participants fixated on to the features of the designated search target using a Siamese Convolutional Neural Network. These cumulation of these features indicate attributes of the target, which can be used to train the support vector machine accurately predict it.