IMU-Based CWT and Whitening-Aided for Human Activity Recognition

In fields such as health monitoring, elderly care, and sports training, accurate identification and tracking of human activities are crucial. If effectively utilized, data collected by wearable and mobile devices can significantly enhance the quality of life and service efficiency. However, traditional time-domain analysis methods struggle to capture the nonlinear and non-stationary characteristics of signals, making it challenging to accurately recognize complex or subtle human movements, thus limiting the adaptability and precision of the models. This paper proposes an approach that converts time-domain signals generated by inertial measurement units (IMU) into richer time-frequency images using continuous wavelet transform (CWT), revealing subtle dynamic features hidden in the raw data. Additionally, the introduced rotation Whitening-Aided process optimizes data representation and reduces redundancy among features, significantly enhancing the generalization ability and accuracy of 98.7% of the recognition models. Experiments conducted on multiple public datasets have validated the effectiveness of our method, which also achieved an accuracy rate of over 94.2%.