Accelerometers

Accelerometers are the most prominent technique for context-aware computing. They are available at low cost (starting at 10 Euros or even less).

Use in current products:

Wii Remote
Some notebook computers (for hard disk protection, but at least on Apple's notebook computers, application software, too, can read the data.)
Digital cameras (automatic rotation of pictures taken in portrait format) [Note that anti-shake requires a gyroscope, however.]

Accelerometers measure (guess what?) acceleration (m²/s), which is the temporal change of velocity (m/s). As is velocity, also acceleration is a vectorial quantity: It is described by a direction and a magnitude or, alternatively, by three components. Very cheap accelerometers can only sense one or two of these components (one-axis, two-axis types); tri-axial accelerometers capture a 3D vector.

Examples for position curves (i.e., motions), velocity vectors, acceleration vectors

Newton: Net applied force is mass times acceleration (F = m*a). No net force, no acceleration and vice versa. From accelerometer data alone you cannot tell an experiment done in a jet plane flying at constant speed from the same experiment done on the ground! Another aspect of this equation: Gravity is a force, so it causes acceleration (downward with the earth's acceleration constant g = 9.81 m/s^2). Thus, we can use an accelerometer to determine the downward direction---as long as there is no other force (or to be precise: as long as all mechanical forces sum to zero) and thus no other acceleration.

Example: ST Microelectronics LIS3L02AQ. How to read the data sheet.

Looking at the ugliness of actual data: tolerances, drift, noise, roundoff, discrete steps in time. A simple way to calibrate an accelerometer is to turn in slowly in all directions and record the minimum and maximum output separately for each of the three outputs. These values should be mapped to -g and +g. Due to the errors, you cannot integrate the accelerometer's output twice to find the position. At least not for more than a couple of seconds. (A Kalman filter may help a little.)

Example: Using an accelerometer to determine the inclination angle and decide when the data is not reliable.

Lab: first steps of pattern classification for accelerometer data; which features to use.