Ground-Based Localization Within Satellite Images

MIT Lincoln Laboratory Computer Science, 2018-19

Liaison(s): Andrew Fishberg ’16
Advisor(s): Zachary Dodds
Students(s): Nupur Banerjee (PM), Arnesh Batlaw, Gary Luo, Jess Winssinger

MIT Lincoln Laboratory is interested in exploring cost-efficient, vision-based localization. This technology is orthogonal to GPS, so it can be used both in place of GPS and as an augmentation to GPS in built environments. Given commodity overhead imagery, such as satellite images from Google Maps or Bing, our team has prototyped and analyzed a system that localizes vehicles using ground-based images. Our system integrates a siamese neural network, particle filtering, and spatially-aware visual features.