jun-geun park

Motion Compatibility for Indoor Localization

We as humans represent their own movement trajectories only by abstract terms describing unit activities or motions, such as "walking", "turning left", "riding elevator up", or "climbing stairs down", associated with, but not necessarily requiring, coarse metric information, such as "for 3 seconds." We can also reconstitute the path taken from such motion descriptors, provided a map with sufficient details. Assume that you can remember the actions you have taken to arrive at your office from the entrance of the building, then you will be able to find the path you have taken from the series of motions.

We explored this idea of finding user trajectory in indoor environment from a series of discrete motions using mobile devices in this work. The method works in two stages: 1) extracting navigational activities from smartphone sensor data, 2) matching the extracted series of motions onto a map parsed automatically from floorplans and represented as a route network to find user trajectory starting from an unknown location. We also devised classification-based algorithms to infer the user's walking speed and the position of the mobile device releative to the user. Here's a youtube video showing how this works in practice.
Relevant publications: 09,10,11

Molé: a Scalable, User-Generated WiFi Positioning Engine

Molé (Mobile Organic Location Engine) is an organic location system, redesigned and extended from our previous design of the organic indoor location system. Its focus is enhanced flexibility and scalability, by adopting hierarchical space representations and cloud-based fingerprint distribution system, among others. See https://projects.developer.nokia.com/mole (project led by Jonathan Ledlie at Nokia Research)
Relevant publications: 06,08

Moving-Baseline Localization

Many algorithms and applications in sensor networks, such as routing or in-network data aggregation, assume the knowledge of sensor locations. The most challenging scenario arises when all nodes are moving in a "GPS-denied" environment, where none of them are aware of their absolute positions. However, most existing work on sensor network localization algorithms assume either that sensors are static or that a fraction of them know their own location. We developed a localization algorithm that is free from such assumptions, which recovers a relative motion configuration, i.e. positions and velocities, of the sensor nodes from pairwise range measurements between them.
Relevant publications: 01,02,03