Runtime Heterogeneous Sensor Selection with Data Reuse in Multi-device Environments

With the increasing deployment of diverse sensory devices on or near a human body, perception resource usage across various applications has been significantly enhanced. However, frequent sensor switching during concurrent multi-sensor operation on mobile devices for different tasks leads to excessive energy consumption. To this end, we propose an energy-efficient data-reuse-based runtime framework called DataReuse, which dynamically selects the best sensor for task execution. It utilizes a single sensor to handle different tasks by sharing data, thereby reducing sensor switching when new tasks arrive. DataReuse includes an adaptive sensor selection mechanism that selects the best sensor based on the current sensor state and the capability values of candidate sensors, and operates under both fixed-period and threshold-triggered policies to ensure optimal performance and timely response. The framework provides a Jacobson/Karels algorithm to evaluate sensor capability and combines it with LFU (Least Frequently Used) to update sensor capability, in order to improve estimation accuracy and reduce unnecessary updates. Our evaluation using the Opportunity dataset shows that DataReuse provides 24% energy savings with comparable accuracy across locomotion and activity recognition tasks when compared to state-of-the-art methods.