An IMU (inertial measurement unit) is a small chip that measures motion. A 6-degree-of-freedom IMU contains an Accelerometer (3-axis linear acceleration) and a Gyroscope (3-axis angular velocity); a 9-DoF IMU adds a Magnetometer (3-axis local magnetic field) for an absolute heading reference. Each axis is one channel, so a 9-DoF IMU emits 9 channels per sample. Some packages also include a barometer for rough altitude, occasionally called a 10-DoF IMU.
IMUs are cheap and ubiquitous — every modern smartphone contains one. Industrial-grade IMUs can cost thousands of dollars; consumer chips are a few dollars apiece. Because the package is small, an IMU can be attached to a person, a vehicle, a robot arm, a drone, or anything else whose motion is worth tracking.
Each sensor alone has limitations. The accelerometer is sensitive to vibration. The gyroscope drifts — small errors in each angular-velocity reading accumulate when integrated, so after a few minutes the integrated orientation is meaningfully wrong. The magnetometer is the most fragile, easily disturbed by nearby iron, magnets, or current-carrying wires.
When all three are fused together — the technical term is Sensor fusion — the IMU can produce a stable estimate of its own orientation in space. The accelerometer senses gravity, the magnetometer senses magnetic north, and together they correct the gyroscope’s drift. This fusion is how a phone knows whether you’re holding it in landscape or portrait without any input.
If we instrument 12 IMUs on a person (one per major segment) we get 108 channels per time sample — a vivid example of the Curse of dimensionality that motivates Feature extraction and Dimensionality reduction.