Transfer Learning Technique for Feature Extraction of Human Activity Recognition
Harish Kumar S R, Kanagasabapathi Somasundaram · 2025
Human Activity Recognition (HAR) is an innovative field with a wide range of applications, including healthcare, fitness tracking, smart home technology, workplace safety, and urban monitoring. In the healthcare sector, HAR aids in remote patient tracking, the early detection of unusual movements, and support during post-operative recovery. Fitness programs utilize HAR to refine exercise routines and monitor activity levels through wearable technology. In smart homes, HAR fosters automation and customization by analyzing the behaviours of residents, while also improving elderly assistance with features like fall detection and emergency notifications. In the realm of workplace safety, HAR is employed to observe employee activities and prevent accidents, and in urban environments, it helps to identify suspicious actions, thus enhancing public safety. This research introduces a comprehensive HAR framework that utilizes Convolutional Neural Networks (CNNs) for effective feature extraction and precise classification of activities from sensor-based time-series data. The model effectively captures both spatial and temporal relationships, ensuring high accuracy and flexibility. Tested against benchmark datasets, the proposed method shows better performance than conventional machine learning approaches. By tackling issues such as sensor data noise and variations in activity patterns, the model is well-equipped for real-time implementation.