Multi-Sensor Data Fusion for Accurate Human Activity Recognition with Deep Learning
Universidad Regional Autonoma de los Andes (UNIANDES), Ecuador, E. J. J. Arias, Universidad Regional Autonoma de los Andes (UNIANDES), Ecuador, Universidad Regional Autonoma de los Andes (UNIANDES), Ecuador, Luz Marina Aguirre Paz, Luis Molina Chalacan
Abstract
In the era of pervasive computing and wearable technology, the accurate recognition of human activities has gained paramount importance across a spectrum of applications, from healthcare monitoring to smart environments. This paper introduces a novel methodology that leverages the fusion of multi-sensor data with deep learning techniques to enhance the precision and robustness of human activity recognition. Our approach commences with the transformation of accelerometer and gyroscope time-series data into recurrence plots, facilitating the distillation of temporal patterns and dependencies. Subsequently, a dual-path convolutional network framework is employed to extract intricate sensory patterns independently, followed by an attention module that fuses these features, capturing their nuanced interactions. Rigorous experimental evaluations, including comparative analyses against traditional machine learning baselines, validate the superior performance of our methodology. The results demonstrate remarkable classification performance, underscoring the efficacy of our approach in recognizing a diverse range of human activities. Our research not only advances the state-of-the-art in activity recognition but also highlights the potential of deep learning and multi-sensor data fusion in enabling context-aware systems for the benefit of society.