Worker Positioning Method Based on Motion-Aware Inertial Regression for Smart Construction Sites

作者：Yujin Kuang, Mujiao Ouyang, Yuan Yang, Xiaoguo Zhang · 年份：2025 · DOI：10.1109/iecon58223.2025.11221223 · 研究领域：Occupational Health and Safety Research、BIM and Construction Integration、Hydraulic and Pneumatic Systems

To address the challenges of high-precision requirements and complex environmental adaptation in worker positioning on dynamic construction sites, this paper proposes a lightweight inertial regression-based positioning method for mobile terminals. By integrating action-aware constraints within a multi-task learning framework, the method enables robust trajectory tracking across various device-holding postures and diverse motion states. It leverages the built-in tri-axial accelerometers and gyroscopes of smartphones to collect inertial data, constructs temporal features using a sliding window approach, and introduces an action-aware velocity regression network. This network jointly learns an auxiliary action recognition task and a primary velocity regression task, enhancing spatiotemporal feature extraction through channel attention mechanisms and residual connections. The incorporation of a multi-task loss function further improves robustness to sensor noise. Experimental results demonstrate that in mixed dynamic environments, the absolute trajectory error of the proposed methods is less than 1.4 meters. It adapts effectively to challenging conditions such as metal occlusion and multipath interference without reliance on external positioning infrastructure. Moreover, it supports intelligent applications including hazardous area warnings and work efficiency analysis. This approach provides a low-cost, highly compatible real-time positioning solution for intelligent constructio...