To solve the location and path planning challenges for intelligent vehicles in GPS-restricted underground parking lots， we employ a vision-based approach that combines binocular offline mapping and monocular online localization to trace the intelligent vehicles within the scene. The improved path planning algorithm is used to plan the global route. First， the system uses a binocular camera to capture scenes at multiple nodes in the underground parking lot， which are then stored in different layers of a hierarchical map. The monocular camera is subsequently utilized to perform coarse-to-fine map matching and calculate the pose of the vehicle， ensuring high-speed and accurate positioning. Finally， the improved path planning algorithm is achieved by detecting intersections and generating a roadmap. The experimental results demonstrate that the average monocular positioning error and average error rate are 1.3 m and 7.4%， respectively， the planning time is reduced from approximately 5 s to less than 0.2 s， and the path length is shortened by approximately 9.56%. This positioning and navigation system is practical for real-time positioning and path planning of intelligent vehicles in parking lot environments.