LiDAR is one of the most widely-used sensors in robot localization. However, the measurement features of optical sensors that occur on the glass surface, such as penetration, diffuse or specular reflections, may lead to unreliable matching between the pre-built map and monitored LiDAR data. Given this context, LiDAR-based localization is affected when applying to a glass-walled environment. To resolve the above problem, this paper proposes an improved ray-casting Monte-Carlo localization (IRC-MCL) method to reduce the scan matching error and enhance the localization accuracy in glass-walled environments. To manifest the environments surrounded by glass walls, a novel grid map with glass information is firstly constructed by obtaining the position and orientation information of the glass walls. Then, based on the measurement characteristics of LiDAR with glass, an improved ray-casting approach is incorporated into the proposed IRC-MCL method to extract the point cloud set from the novel grid map, which is used as the input of the scan matching to improve the matching precision. The feasibility of the proposed IRC-MCL method is verified through several real-world experiments, and the localization accuracy is guaranteed to be within 0.005 m/0.05 deg in the environment surrounded by glass walls.