Exploring Human Activity Patterns: Investigating Feature Extraction Techniques for Improved Recognition with ANN

In recent decades, human motion recognition has experienced significant expansion which is primarily due to advancements in sensor technology and deep learning. In this paper, we propose a human motion capture and recognition approach based on the Arduino BLE 33 Sense—a compact and lowcost development board equipped with an Inertial Measurement Unit (IMU) sensor. Our objective is to accurately classify five distinct types of human motions: walking, running, ascending or descending stairs, remaining in a steady state, and using a lift. In order to pre-process our data, we used two feature extraction techniques: Fast Fourier Transform (FFT) and Wavelet. We classify and predict human motion using an artificial neural network architecture. The experimental results shows that our classification model boasts an impressive accuracy rate of 94% based on the FFT feature extraction technique.