The use of three-dimensional point cloud data for road information acquisition is very important to improve the intelligence level of unmanned fire-fighting vehicles used in warehouses. In the functional architecture of unmanned fire-fighting vehicles, obstacle detection plays an important role, which can improve the efficiency of unmanned fire-fighting vehicles and realize the timely extinguishment of fire sources. This paper analyzes and summarizes the obstacle detection algorithm based on mechanical LiDAR point cloud data. Aiming at the real-time requirement of obstacle detection and the problem of point cloud data distortion caused by the movement of unmanned fire-fighting vehicle, the running speed of the algorithm is improved by point cloud preprocessing, and the distortion correction algorithm is used to reduce the distortion caused by the movement of the vehicle. The traditional clustering algorithm is improved, which can detect close or far obstacles accurately and reduce the influence of noise and outliers on the clustering effect. The algorithm is applied to the processing of the original data set and the effectiveness of the algorithm is evaluated. The results show that the obstacle clustering accuracy is increased by 5.1%, and the clustering speed is increased by 4.9%, which can realize the accurate and rapid detection of various obstacles, and has certain guiding significance for the improvement of the obstacle avoidance ability of unmanned fire trucks used in warehouses.