We propose a partial localization approach for cloud-based visual localization. By applying geometric lifting to the map we prevent the server from knowing the exact client pose. Still, the client can locally combine multiple partial queries into a full pose.
We present privacy preserving localization and mapping without the need for calibrated cameras.
A method for automatic camera rig calibration based on a prebuilt model of the environment.
We present the first full Structure-from-Motion pipeline from privacy preserving line features.
A novel, dynamic feature matching and pose estimation strategy tailored to multi-camera systems.
A project to develop visual perception systems for autonomous driving in difficult environments.
Direct VIO for multi-camera systems.