In the last years multiple works have used lifting of geometric features as a way of protecting privacy in cloud-based visual localization. These works aim to prohibit reconstruction of the scene appearance from the shared data.
In this project we apply a similar technique to hide a client's exact pose from the localization service. We achieve this by adding additional degrees of freedom to our map representation so that the client's position can only be estimated along a single direction.
The client then queries multiple independent servers with different maps to obtain constraints in different directions, and only combines the results locally into a full pose.