Database Reference
In-Depth Information
sensor technology. The lab adopts these technologies for the purposes
of retraining gait in children with cerebral palsy and is leading research
into development better prosthetics for amputees, interactive technol-
ogy for stroke survivors, and traumatic brain injuries and people with
burn-related contractures.
There is emerging interest in building Body Area Sensor Networks
- large-scale BSNs across a public healthcare system such as a hospi-
tal. The miTag system [95] is a pilot public healthcare BSN deployed
in the Baltimore Washington Metropolitan region. This system includes
a wireless multi-sensor platform that collects information from GPS re-
ceivers, pulse oximeters, blood pressure cuffs, temperature sensors, and
ECG sensors. The system supports two-way communication between
patients and healthcare providers, to allow for feedback based on the
monitored health and context information. Body Area Sensor Networks
are also being developed to support disaster management in emergency
response systems.
The maturity of sensor networks has allowed the development of smart
environments for wellness and chronic disease management. For exam-
ple, some researchers have used smart environments with combinations
of wearable devices (RFID bracelets) and RFID tagged objects to de-
tect indications of cognitive impairments such as dementia and traumatic
brain injury (TBI) by monitoring individuals performing a well defined
routine task - making coffee [103]. The researchers define and compute a
set of four domain specific features from the sensor data, that are increas-
ingly representative of the task, and correlate with severity of cognitive
impairment. These features include the Trial Duration, Action Gaps,
Object Misuse, and Edit Distance. Trial Duration captures the total
time taken for the activity while Action Gaps represent periods during
which subjects were not interacting with any objects on the assumption
that during those periods they are considering what step to take next.
Object Misuse captures the number of times a subject interacts with
each object used in the task - with failure to interact with a required ob-
ject, or an excessive number of interactions indicates problems. Finally,
the researchers manually define a representative plan 5 for the task, that
represents a partial order (to allow alternate reasonable task executions)
over object interaction. The Edit Distance, as used in natural language
processing then captures deviations from this plan. Finally, these fea-
tures are analyzed using Principal Component Analysis (PCA) to ex-
amine correlations between computed features and larger trends in the
5 Other research on activity recognition has addressed the question of automatically con-
structing plans for everyday activities by mining the web for descriptions of these activities.
Search WWH ::




Custom Search