The Application of Autoencoder In IMU Data Process

Inertial Navigation System (INS) and Inertial Measurement Unit (IMU) are commonly used technologies in modern navigation and attitude control fields. INS is a navigation system based on the principle of inertial sensors, which calculates the position, velocity, and attitude of an object in real-time by measuring its acceleration and angular velocity and using initial attitude information. IMU is the core component of INS and usually contains three-axis accelerometers and three-axis gyroscopes. Accelerometers are used to measure linear acceleration, and gyroscopes are used to measure angular velocity. In INS and IMU, sensors usually collect the measurement values of multiple accelerometers and gyroscopes at high frequency. These measurement values contain a large amount of information, including object acceleration, angular velocity, and attitude changes. However, due to the large number of multi-axis sensors and high-frequency sampling, processing and analyzing these data may become complex and time-consuming. At this time, data dimensionality reduction techniques can be applied[1]. The main idea of this article is to add residual blocks to the traditional dimensionality reduction technique autoencoder to improve the model's ability to extract important features. This approach can transform high-dimensional sensor data into lower-dimensional representations while retaining essential motion features. This simplifies subsequent data processing, analysis, and storage processes, improves computing efficiency, and saves resources.