What is IMU?
IMU (Inertial Measurement Unit) is one of the common sensors to provide motion data in a time-series format. In this post we review it.
IMU (Inertial Measurement Unit) sensor provides time-series data, used in Human Activity Recognition problems, Tracking & Navigation problems, and many more. In the AI era, this cheap and reliable sensor can provide a lot of data, so one might generalize it using machine/deep learning to many useful insights in a various range of fields.
Where can we find an IMU sensor?
Actually, almost everywhere! Let’s start with our smartphones. Most smartphone devices are equipped with an IMU sensor inside, a MEMS (Micro-electromechanical Systems) technology. It is also placed on many Tablet devices. The IMU is very common in the automotive and aerospace industries, as it allows a better determination of the vehicle's position and orientation.
The Inertial Measurement Unit is one of the common sensors in the navigation field (but not only). It contains an accelerometer and gyroscope (sometimes also a magnetometer and rarely also a barometer). The first is responsible for acceleration measurements and the second is responsible for the angular velocity measurements. Each one of the measures is represented in a three-axis coordinate system, so generally speaking, they both together yield a 6-dimension measurement time series stream.
How does it work?
So, let’s divide the discussion into 4 parts against 4 different sensors:
1. Accelerometer is a sensor that measures the specific force (the body mass normalizes the force). It provides the acceleration across the x, y, and z axes in its local frame.
2. Gyroscope is a sensor that measures angular velocity around the x, y, and z axes, in its local frame. Generally, integrating the measurements results in the angles themselves.
3. Magnetometer is a sensor that measures the Earth’s magnetic field and provides the heading (the compass is one such device). If it is included in the IMU, we commonly describe it as a “9-axis IMU.”
4. Barometer is a sensor that measures air pressure and can provide altitude.
Tracking Relation and Major Disadvantages
One of the major IMU’s disadvantages is their accumulated error when integrating them with navigation equations alone in order to find the position. For example, double integration of the noisy acceleration leads to an inaccurate position solution that eventually results in a huge drift. Hence, many tracking/navigation applications use additional sensors, such as GNSS receivers and cameras. These sensors provide position information at a lower frequency and provide resetting the accumulated errors.
The HAR problem is a great use for the IMU data
One of the famous time-series datasets is the Human Activity Recognition which contains recorded IMU signals, released in 2014. A collection of some activities was tagged and stored at 50[Hz] frequency. It is widely used to evaluate state-of-the-art time-series classification algorithms. To read more, you may refer to the original paper. Many algorithms were established to recognize the human activity based on the IMU signals only, as it can be formulated as a classification problem. The major advantage of using IMU signal for HAR over images/video, etc., is the small size of data. This advantage allows a compact representation of the HAR classifier.