Fisheye-Camera-Based NLOS IS-VLP with Distortion-Aware Geometric Modeling

This letter proposes a fisheye-camera-based non-line-of-sight image-sensor-based visible light positioning (NLOS IS-VLP) system for wide-field-of-view indoor localization, using distortion-aware geometric modeling. By establishing a camera-ground-virtual-LED spatial geometry, the proposed method derives the NLOS VLP formulation under a fisheye distortion model. Floor-reflected bright spots are modeled as virtual LED projections, and a high-order polynomial inverse distortion process maps pixel observations to spatial directions. The receiver position is then estimated by minimizing the reprojection residual. Experiments show that the proposed system achieves an average positioning error of 13.89 cm and a root mean square error of 14.81 cm, with the 90th-percentile error below 21 cm. These results demonstrate that the proposed model outperforms state-of-the-art NLOS IS-VLP methods while providing a 180° field of view without requiring auxiliary sensors.