Hand Hygiene

We wondered whether it would be possible to detect a handwashing behavior with a wrist-mounted accelerometer. We bought an accelerometer, and did some preliminary testing. The accelerometer takes an input voltage of 2.2v to 3.6v. An input voltage of 3.3v was used for our testing. It has three output voltages which correspond to the accelerations on the x, y, and z axes.  The max/min acceleration can be set to +/- 1.5, 2, 4 or 6g (g being the acceleration due to gravity). For our test, this was set to +/- 1.5g. An output voltage of 0v means that axis is experiencing minimum acceleration. Zero acceleration occurs between 1.5 and 1.8v, and maximum acceleration occurs between 2.3 and 2.6v. All this data and more is available on the data sheet.

The accelerometer was attached to the wrist while performing various motions, and the output voltages were graphed over time.

Here is a graph when the accelerometer is standing still on the table top:

As you can see, there is a notable amount of noise, although the signal is still visible. The z axis is experiencing around 1g of acceleration since it is nearly parallel to the force of gravity, whereas the other axes are nearly perpendicular to the force of gravity so they are experiencing around 0g of acceleration.*

Here is a graph of the output voltages during vigorous hand washing motions:

Here we can see somewhat periodic fluctuations in the voltages. The output for less vigorous hand washing looks like this:

Here the fluctuations are less pronounced, but we still see a periodic trend (perhaps even more so). This information lends the accelerometer to use in the hospital. Accelerometers of this type could be put into wristwatch devices worn by doctors. After a puck is pressed, it could listen for a signal from one of these wristwatch devices, and if the trends in acceleration over a certain period of time match those of hand washing the puck would confirm that a doctor has washed his/her hands. In order for this to happen, however, a hand washing motion must have sufficiently different output from other motions the doctor may be doing.

Here is the output when the subject is idly swinging their arm:

Here we see a somewhat less pronounced periodic trend.

Here is the output when the subject is making emphatic gestures while talking:

This has potential to be confused with a hand washing motion since the magnitude of the fluctuations is similar to that of vigorous hand washing. This is particularly true if the gestures are repetitive in nature, resulting in a periodic graph. This case may be rare enough that it can be ignored.

Altogether, there are several things that must be dealt with for this application to be put into effect. For one thing, the motes would not fit in a wristwatch-size device. We would need to find something else which performs the same function as a mote, or connect the accelerometer to a pager through a chord. We would also need to figure out how to analyze the data quickly in the hospital. The setup used for this experiment involved one data point every 2-3 ms, resulting in about 10 kb/s of data flow.

At this point, more tests should be done on more people to determine if a hand washing motion has a predictable enough output. One possibility would be to put the accelerometer in a wristwatch device connected to a pager with a chord, and have a flash drive connected to the mote in the pager to store the data. This would allow us to collect data from the hospital to be analyzed at a later time.

*I use acceleration and gravity interchangeably here since we know from the general theory of relativity that the accelerometer won’t be able to tell the difference.