Mercury: A Wearable Sensor Network Platform for High-Fidelity Motion Analysis

Mercury is a sensor network platform designed to supports applications that are data-intensive and can gracefully adapt to fluctuations in resource availability and load. Key challenges addressed by Mercury include long sensor node lifetime, autonomous operation, and the need for the system to automatically tune its behavior in response to fluctuations in radio bandwidth and energy availability.

We are currently working with the Motion Analysis Laboratory at Spaulding Hospital to develop a prototype of this platform for long-term motion analysis studies in a home setting. A Mercury network consists of a number of wearable sensors and a base station installed in the patient's home. Each sensor samples multiple channels of accelerometer, gyroscope, and/or physiological data and stores raw signals to local flash. Sensors also perform feature extraction on the raw signals, which may involve expensive on-board computation. The body sensor network performs opportunistic data transfer to the base station, based on the quality of the radio link to each sensor and the remaining battery capacity. Each node dynamically tunes the number of data transfers and degree of computation applied to the sampled signal to meet a target lifetime (say, 12 or 24 hours). Nodes also save energy by dropping down to a low-power state when the sensor is not moving.

Click here for papers related to Mercury

An earlier version of Mercury (v1.0) is being used in several studies by the Motion Analysis Lab at Spaulding Hospital.

Parkinson's disease
Clinical evaluations of patients in various settings, including tuning of deep brain stimulation (DBS) parameters. Patients are brought in to the lab for a 4-hour session. Every 30 min, patient is asked to perform various movements and activities, for about 10 min. Data collection is performed manually by clinician in the lab using Mercury v1.0. There are 9 nodes on the body (2 on each arm and leg, one on the back), recording triaxial accelerometer and gyroscope data at 100 Hz. (As of 7/24/08: 4 patients, 7 sessions each, 4 hours per session.)

Epilepsy
This is just starting up. The goal is to detect seizures, which are often subtle and difficult to detect. The number of sensors varies because it depends on type of seizures the patient has. In a typical setup there are 4 sensors on arms and only 2 on legs. Accel, gyro, and EMG data is collected. EMG is be localized (say, one arm) and sampled at 500 Hz. (As of 7/24/08: 2 patients, measured for 1 week each in hospital.)

Hardware platform details

We are using a range of wireless sensors in this project. Our current hardware platform is the SHIMMER wearable mote, developed by the Digital Health Group at Intel. SHIMMER incorporates a TI MSP430 processor, CC2420 IEEE 802.15.4 radio, triaxial accelerometer, and rechargeable Li-polymer battery. SHIMMER includes a MicroSD slot supporting up to 2 GBytes of Flash memory. This allows SHIMMER to store significant amounts of data (2GB can store more than 80 days of continuous triaxial accelerometer data sampled at 50Hz). SHIMMER can also be configured with an optional Bluetooth radio.

SHIMMER devices are now available commercially from www.shimmer-research.com.


The Intel SHIMMER mote, including a triaxial accelerometer.	The SHIMMER mote connected to its programming board.	SHIMMER sensors being worn on a patient's arm.

Our previous hardware platform is the Pluto mote, designed here at Harvard, which is a scaled-down version of the Telos designed to be small, lightweight, and wearable. The Pluto incorporates a tiny, rechargeable Li-ion battery, small USB connector, and 3-axis accelerometer.