Latency-Aware Construction Activity Recognition Using Kinematic and Physiological Sensors

Abstract: Wearable sensors are increasingly being used to monitor worker activity and safety. While Human Activity Recognition (HAR) studies predominantly rely on kinematic data, physiological signals remain mostly underexplored due to the delayed nature of physiological responses. This study investigates the fusion of physiological and kinematic modalities by systematically evaluating temporal offsets to compensate for physiological latency. Results show that classification performance is sensitive to delay magnitude, with optimal performance achieved using an 8-second window and a 4-second time lag. Based on these findings, a latency-aware fusion protocol is proposed to enable robust multimodal HAR in wearable-based worker monitoring systems.