195 / 2023-09-07 16:21:59

Multi- Source Fusion Techniques for High-Precision Indoor positioning in Complex Subterranean Environments

Indoor positioning; Multi-source Fusion; Generative Adversarial Neural Networks(GANN); Gated Recurrent Unit(GRU); Factor Graph

    Ensuring the personal safety of underground workers in emergency situations is a crucial prerequisite for industrial precision mining operations. To address the personnel localization requirements in complex underground environments, we propose a multi-source fusion method based on deep learning, integrating geomagnetic data, ambient light, and Pedestrian Dead Reckoning (PDR). To minimize human and material resource consumption, we construct a Conditional Generative Adversarial Network (CGAN) model during the offline phase. By leveraging a limited set of known samples, the model estimates the fingerprint strengths for uncollected areas, thereby building a high-resolution and high-stability geomagnetic/ambient light fingerprint database. In the online phase, we employ a Tent-chaos-mapping-optimized Sparrow Algorithm (Tent-SA) to refine a Gated Recurrent Unit (GRU) model. This model outputs positioning results that fuse geomagnetic and ambient light data.

    To further enhance accuracy, we also design a sophisticated cascading filter architecture that combines the strengths of adaptive Kalman and Unscented Kalman Filters (UKF). By introducing a sliding-window mechanism, our solution innovatively adapts the environmental noise covariance parameters. This not only effectively mitigates the problem of cumulative trajectory drift in traditional PDR-based positioning systems but also significantly boosts the system's overall reliability. Ultimately, we leverage factor graphs to fuse the improved PDR tracks with the geomagnetic/ambient light positioning paths, further enhancing the location accuracy and stability of personnel in complex underground settings. Moreover, in scenarios burdened with high magnetic interference or unstable ambient light conditions, our system offers the flexibility to switch between geomagnetic/PDR or ambient light/PDR combination positioning methods. This adaptability makes it invaluable for emergency coarse-grained positioning. Experimental results on our self-collected dataset demonstrate that compared to traditional localization methods, the approach presented in this paper offers excellent positioning performance and versatility while maintaining low costs.