External Reflection Recognition for Glass Walls Surface

Abstract

Most buildings in modern cities contain many glass facades to show the modern flavor of architectural design and street view, gradually increase the intensity of the lighting, which is also the first choice for most green and energy-saving buildings. However, the strong reflectivity of the glass facade allows street view or sky to be reflected on the surface of the building, while the transparency allows the visualization of internal objects and interferes with the monitoring of the glass building itself. Such features increase the difficulty of urban building monitoring. In this paper, a multi-layer-separation algorithm based on semantic segmentation method is proposed to realize non-contact monitoring of glass buildings, which separate the reflection and the real object from the monitoring images.