Smartphone-based context-aware system for human activity recognition: dynamic and static methods

Smartphone-based context-aware system for human activity recognition: dynamic and static methods

연구 분야: Strategies

학회: Multimedia Tools and Applications

Human Activity Recognition (HAR) using smartphones has gained increasing prominence in several fields such as healthcare, security, and surveillance. These systems leverage sensor data from smartphones, and often contextual information (e.g., location, time), to accurately recognize activities. However, existing HAR methods typically rely on the full set of available sensor and contextual data, even for simple activities, leading to unnecessary computational costs and resource usage. In this context, the paper presents two novel machine learning-based contextual HAR methods: a Static Context-aware System for HAR (SCSHAR) and a Dynamic Context-aware System for HAR (DCSHAR). While SCSHAR uses all available sensor and contextual informations simultaneously for recognition, DCSHAR dynamically selects only the necessary subset of sensor and contextual information, reducing resource consumption without compromising accuracy. The proposed methods were evaluated on three publicly available datasets, DOMINO, ExtraSensory, and RealWorld, and compared with state-of-the-art approaches. Results show that SCSHAR achieves slightly better performance than competing methods, while DCSHAR, despite using less information, achieves performance comparable to, or even superior to, that of SCSHAR in most cases, demonstrating its efficiency. For instance, DCSHAR achieves 91%, 97%, and 99% accuracy on the ExtraSensory, DOMINO, and RealWorld datasets respectively, with corresponding macro F1-scores of 75%, 96%, and 98%. Moreover, DCSHAR reduces execution time compared to static approaches, making it suitable for real-time applications.