A Wearable Interface for Topological Mapping and Localization in Indoor Environments

G. Schindler, T. Starner, and C. Metzger. A Wearable Interface for Topological Mapping and Localization in Indoor Environments. 2nd International Workshop on Location and Context-Awareness (LoCA), 2006.

The authors present a method for mapping and localization in indoor environments using FreeDigiter, an ear-mounted gesture interface equipped with an infrared proximity sensor (to detect footsteps, doorways and finger gesture) and a dual axis accelerometer. The mobile robotics community has studied the automatic mapping of unknown indoor environments. Normally, in location-recognition works with wearable accelerometers a dead reckoning approach is used. The accelerometers data are integrated over time to build a metric map of a user’s path through an environment. FreeDigiter also takes the user’s steps, but also captures the connectivity of an indoor environment composed of multiple rooms. It builds and tracks a topological map. In robotics literature focuses on Simultaneous Location And Mapping (SLAM). In their projects, the authors propose a cheap, lightweight device requiring minimal user intervention.

Mapping and localization is done by first building an augmented topological map:

A map is represented as a set of edges E and vertices V defining a graph G = {E, V }. Each edge is augmented with a length l (in footsteps) and an edge-specific probability distribution over proximity sensor readings (the mean μ and variance σ of a Gaussian). By this definition, each edge corresponds to a constant part of the environment – i.e. the world looks the same to the sensors at every point on an edge.

In the building, the proximity sensor is used to detect and measure doorways and use an accelerometer to determine the distance between doorways. Once the map is constructed, the autors used a particle filter to track the user’s movements accross the edges of the graphs. With an experience FreeDigit user tracking accuracy reach 100%. This means a user has to be able to maintain a constant speed and be a consistent walker.

Relation to my thesis: A lab experiment using mapping and localization for person tracking in unknown indoor settings. The authors the knowledge in robotics to person tracking. It is not a surprise to see that somehow the users must behave like robots (keep pace and consistent) for the system to be performant (adding more sensors can alleviate the problem). However, I find interesting that the user takes part of the mapping and trains the location application. The user helps disambiguates. I was not aware of SLAM and might find there some relations between robot and person localization and it provides good reference in robotics localization:

On partical filtering:
S. Thrun, D. Fox, F. Dellaert, and W. Burgard. Particle filters for mobile robot localization. In Arnaud Doucet, Nando de Freitas, and Neil Gordon, editors, Sequential Monte Carlo Methods in Practice. Springer-Verlag, New York, January 2001.

On Voronoi tracking
L.Liao, D.Fox, J.Hightower, H.Kautz, and D.Schulz. Voronoi tracking: Location estimation using sparse and noisy sensor data. In IROS, 2003.