Abstract: A glass surface detection method based on visual distortion clues is proposed to address the challenges in detecting glass surfaces caused by the characteristics of glass materials. Firstly, the backbone features are processed by a visual distortion aware module. By leveraging the phenomenon of visual distortion in the image regions covered by glass, the backbone network is guided to perform an initial localization of the glass surface, thereby obtaining an initial glass feature map. Subsequently, a structural refinement module is employed to progressively refine the initial glass feature map by utilizing the information about the number of glass objects in the image, resulting in finely-detailed edges of the glass surface. The experimental results conducted with 17 different methods indicate that the proposed method achieves improvements in IoU ranging from 0.83% to 4.73%, improvements in F-measure ranging from 1.40% to 6.60%, reductions in MAE ranging from 0.72% to 2.60%, and decreases in BER ranging from 0.58% to 2.66% across four benchmark datasets.