Frederike Dümbgen, EPFL

Abstract

Humans use a wealth of heterogeneous signals to navigate through space. We leverage the rich visual signals from our retina, and combine it with acoustic and inertial measurements from the auditory system, tactile signals, and others, to sense our surroundings. We learn to travel efficiently through an environment, to reorient ourselves when we get lost, and to create cognitive maps for future use. The diversity of signals improves the accuracy of our navigation, and it provides us with a resilience to failure of the underlying systems.

Inspired by this, we develop solutions which aim at providing accurate localization by aggregating signals from different modalities. In this talk, I will provide both theoretical and practical insights which can pave the way for such multimodal systems. I will present Coordinate Difference Matrices (CDMs), which we developed to solve various problems arising in localization and mapping, for example localization from ranges and angles. Then, I will present our ongoing research on the reconstruction of parametric trajectories, which eliminates the necessity for dense and synchronized distance measurements. Finally, I will share results from our real-world indoor localization system combining Bluetooth, WiFi, inertial measurement units and visual measurements.

Part of my presentation contains research done by my colleagues from the laboratory of audiovisual communications (LCAV) or in collaboration. The last project is a collaboration with the school of engineering and architecture of Fribourg (HEIA-FR) and Vidinoti.

Video