Final Examination of Krishna Regmi for the degree of Doctor of Philosophy in Computer Science
Download Bifold
We have developed deep neural network based architectures to generate cross-view images, e.g. aerial image corresponding to the ground-level scene, and also developed a method to subsequently utilize the features of synthesized images to facilitate geolocalization task employing image-based matching. Such localization is useful in vision based navigation in a GPS denied environment, e,g. a moving car can selflocalize itself in case of GPS failure. Similarly, given the aerial images of a particular region in the world over time, we can analyze the changes (development or destruction) the region undergoes and study its historical evolution. Additionally, the authenticity of the geo-tagged images/videos uploaded in the social media can be validated, which can help purge the fake information from being circulated in the media.
Read More