ABC-iVLP: Indoor Visible Light Positioning Using Attention Mechanism-Based BiLSTM-CNN

Indoor positioning is essential in applications such as intelligent warehousing and innovative healthcare. Conventional solutions, such as the Global Navigation Satellite System (GNSS) and Wi-Fi, suffer from low accuracy or high deployment costs. In contrast, visible light positioning (VLP) offers immunity to electromagnetic interference and low-cost deployment. However, VLP systems still face challenges due to multipath-induced noise and insufficient use of temporal correlations in fingerprint models. To address these issues, an attention-mechanism-based bidirectional long short-term memory (BiLSTM) and a convolutional neural network indoor VLP (ABC-iVLP) are proposed in the paper. The proposed ABC model comprises a three-stage framework for dynamic weight allocation that incorporates a signal-to-noise attention mechanism to dynamically suppress noise, a BiLSTM to capture temporal dependencies in received signal strength, and a Convolutional Neural Network (CNN) to extract spatial features. Additionally, the proposed ABC-iVLP model is validated in a 2 m × 2 m × 3 m room, yielding a maximum positioning error of 6.47 cm and an average positioning error of 2.33 cm. Compared with the ABC iVLP model without attention, BiLSTM, and CNN, the average positioning error is reduced by 32.5%, 43.4%, and 18.7%, respectively. Furthermore, robustness is verified across varying LEDcounts and ground heights, demonstrating the superiority of ABC-iVLP over existing schemes.