Research project title:
Visual Navigation Using Computer Vision and Deep Learning
Th e first part of my PhD project uses computer vision and deep learning to learn weather and lighting invariant representations of place images and combines this compact representation with visual odomoetry in a novel implementation of a particle filter to enable route navigation that exceeds the current state of the art by operating in real time and being robust to large visual changes.
The second part uses a convolutional triplet network to identify user-defined places of interest along a route. Places of interest detection is then combined with memorability and distance sampling to compress a visual map while maintaining localisation accuracy at those places of interest.