Glass walls detection and reconstruction method based on Fresnel optical model and GBR-RANSAC algorithm for indoor environments

Abstract Glass walls are widely utilized in modern architecture due to their aesthetic and functional benefits. However, their unique optical properties, including reflection, transmission, and low reflectivity pose significant challenges for automated guided vehicles (AGVs) relying on LiDAR-based environmental perception. The presence of glass walls can severely distort or occlude LiDAR measurements, degrading the integrity and precision of environmental maps and escalating collision risks for AGVs operating near these transparent barriers. To address these challenges, this study proposes a novel method for glass wall detection and reconstruction by integrating Fresnel optical modeling with the gradient boosting regressor-RANSAC (GBR-RANSAC) algorithm. First, we establish a nonlinear mathematical relationship between reflection intensity and incident angle through Fresnel equations and Gaussian distribution, providing a theoretical foundation for precise glass wall detection. Second, the proposed GBR-RANSAC algorithm enables accurate position calibration and is integrated into the LIO-SAM framework to achieve robust 3D reconstruction. Experimental results demonstrate exceptional performance: the method achieves an average positioning error of 1.81 cm, a maximum error of 2.13 cm, and an outlier rejection rate of 97.63% across varying distances and incident angles. Furthermore, the algorithm processes point cloud data in 9.7 ms per frame, meeting real-time requirements for indoor navigation systems.