Human Activity Recognition via Time Series Imaging of Wearable Sensors and Fusion of Multiple Deep CNN Models
Mohammad Sakka, Osama Orabi, Yazan Alnakri, Mohammad Reza Bahrami · 2025
Human Activity Recognition (HAR) is essential in healthcare, smart homes, and security applications. Recent advancements in sensor technology and machine learning have enhanced HAR's ability to address real-world challenges, including continuous monitoring of elderly individuals. This study presents a HAR system developed using the publicly available GOTOV dataset, which contains accelerometer data collected from the ankle, chest, and wrist. The time-series signals were converted into images using the Gramian Angular Field (GAF) technique, allowing Convolutional Neural Network (CNN) models to be trained for each sensor modality. To improve classification performance, a fusion strategy was employed to integrate the predictions from all sensor-specific models. The proposed system achieved an F1-score exceeding 93% across 16 activity classes, demonstrating its effectiveness in multi-class human activity recognition.