Visual Localization based on Binary Features

Recently, Google, Microsoft and several start-ups have started to launch services for indoor maps. Due to its potentially high localization accuracy and its independence from hardware installations, visual indoor localization and navigation for hand-held devices is becoming a hot topic. A visual localization system consists of a visual odometry system with an integrated relocalization algorithm and a global localization mechanism for initialization. The deployment of feature based pose recovery algorithms on hand-held devices has mostly been avoided, due to computational complexity of the feature descriptors. Binary features like Binary Robust Features (BRIEF) promise to overcome this problem as they are about 40 times faster to extract than the quasi standard descriptor SURF. At the same time they offer comparable matching precision under small rotations and scale changes. This thesis investigates in the deployment of binary features for global pose recovery as well as relocalization within the visual odometry system. Integrated in the Parallel Tracking and Mapping (PTAM) algorithm, the developed BRIEF based relocalization algorithm was found to yield accurate, fast and robust pose recovery even in sparsely textured and repetitive indoor environments. For global localization, a novel quantizer for binary features was developed to enable Content Based Image Retrieval (CBIR). In combination with a Virtual Views database the high distinctiveness of BRIEF features can thus be leveraged to perform accurate global visual localization. This large scale visual localization system matches the precision of state of the art SURF based system, while reducing the computational burden of feature extraction significantly. To conclude, deploying binary features for pose recovery leads to significant speedups in the feature extraction without loss in localization performance. This makes binary features ideal for mobile-device visual localization systems.

I wrote my thesis in the NAVVIS group of the Institute for Media Technology at the Technische Universität München.

Bibtex

@mastersthesis{straub2012diplomathesis,
  author = {Straub, Julian},
  title = {Visual Localization based on Binary Features},
  school = {Technische Universi{"a}t M{"u}nchen},
  month = {Aug},
  year = {2012},
  address = {M{"u}nchen},
  url =
  { http://people.csail.mit.edu/jstraub/download/straub2012diplomathesis.pdf}
}