Guide dogs can assist visually impaired individuals in avoiding obstacles and reaching their destinations with high training costs. Compared to guide dogs, quadruped robots offer advantages like lower costs, high traversability, and the potential for mass production. In order to complete guiding tasks safely, robots must perform precise localization and terrain reconstruction for path planning and foothold determination. However, current perception systems cannot match the mobility performance of guided quadruped robots, especially in challenging terrains. Considering the size and payload limitations of guided quadruped robots, this paper presents a multi-sensor fusion perception system, composed of localization and mapping modules, that enables guided quadruped robots to perform guiding tasks autonomously. The localization module employs a factor graph-based LiDAR-inertial odometry for global positioning, while the mapping module utilizes camera point clouds to reconstruct local terrain. The proposed method was validated through a simulation study to reconstruct typical terrains, where its real-time performance, robustness, and versatility are characterized.