Improving Lower Limb Activity Recognition Based on Inertial Sensors Using CNN-LSTM Network

The utilization of wearable sensors has gained significant popularity across various domains. This widespread adoption has driven the need for advancements in sensor technology, focusing on attributes such as lightweight design, dependability, and integration capabilities. One specific area where wearable sensor networks have faced challenges is in human activity recognition, particularly in accurately inferring movement data. The effectiveness of this process depends on several factors, such as the quality of sensor-derived features, the discriminatory power of selected features, and the chosen classification methodology. To address these challenges, this research introduces a novel hybrid deep learning architecture called CNN-LSTM. The primary goal of this architecture is to enhance the accuracy of recognizing human lower limb movement. To evaluate the effectiveness of the proposed model, a series of investigations were conducted using the publicly available HuGaDB dataset. This dataset consists of data collected from a body sensor network comprising six wearable inertial sensors, including accelerometers and gyroscopes. These sensors were positioned on both the right and left thighs, shins, and feet. The outcomes of the investigation demonstrated that the suggested model achieved an impressive precision of 97.81 % and a maximum Fl-score of 95.50%. These results highlight the effectiveness of the CNN-LSTM model in accurately recognizing human lower limb movement.