Location is a key information for context-aware systems. While coarse-grained indoor location estimates may be obtained quite easily (e.g. based on WiFi or GSM), finer-grained estimates typically require additional infrastructure (e.g. ultrasound).
This project explores an approach to estimate significant places, e.g., at the fridge, without requiring infrastructure. We use a pocket-based iner- tial measurement sensor, which can be found in many recent phones. We analyze how the spatial layout such as geographic orientation of buildings, arrangement and type of furniture can serve as the basis to estimate typical places in a daily scenario.
Spatial constraints in our daily life
The environmental layout predefines potential places that we can reside in. The layout of buildings, the arrangement and type of furniture or significant landmarks (such as billboards) structure the space in which we are active. For example, when a person is sitting at her working desk she typically faces a certain orientation (Fig 1 and Fig 2). Likewise transiting between different places is characterized by a specific orientation trace as our movements are constrained by walls, doors, stairs, etc (Fig 1 and Fig 3). The blue arrow illustrates a place transition characterized by a sequence of orientation changes.
Using human orientation seems intuitively promising for place detection and thus location estimation.
Fig 1. Typical facing orientation in daily living constraint by spatial environment (red)
Detecting places and place transitions [1,2]
We present a method that segments continuous data into static places and into in-transit segments between places, and matches unknown places to a set templates of typical places. Inertial Measurement data is collected using an Xsens IMU, worn in the pocket.
Using a joint model that combines places and place transitions, we are able to discriminate
nearly 40 places at high accuracy.
Handling displacement [2]
The phone - as sensor platform - can be put in various ways into the pocket. Furthermore, the orientation is often subject to change during walking. By using, a calibration method based on specific leg movement (see Fig 4), the orientation of the sensor is estimated with respect to the body. Hereby we are able to reliably detect location transitions in a office scenario.
Fig 2. Example for orientation information for typical daily places (left and middle). Sequence of absolute orientation
while transiting between places (right).
Fig 4. Principle component on
gyroscope values while walking
