Adversarial Learning for Visual Tracking Research Idea

The doctoral research activity1 mainly focuses on methodologies in the field of computer vision. In particular, the work is focused on designing, developing and validating novel approaches, also based on deep learning methodologies, for visual tracking. Visual tracking in video sequences has always been a main topic in computer vision and interesting results have been obtained by approaches based on Support Vector Machine, Siamese Networks and Discrete Correlation Filters. However, these techniques are limited due to the low discriminative ability of the used features for object detection. In his research activities, Emanuel Di Nardo proposes a novel approach, based on Generative Adversarial Networks for feature extraction or regression. In particular, using Generative Adversarial Networks we are able to characterize the elements to be traced in the scene and make them easier to recognize.