To address the problem of accuracy degradation and poor system robustness of laser SLAM in a dynamic environment, a laser SLAM algorithm for unmanned vehicles based on a depth map in a dynamic environment is proposed. Firstly, aiming at the problem that some ground point clouds are misjudged as dynamic obstacles when the laser incident light is close to the ground, the laser point cloud is divided into ground point cloud and non-ground point cloud by angle threshold, excluding the influence of ground point cloud. Secondly, the non-ground point cloud is projected into a depth map, and the point cloud in the 3D is projected onto a two-dimensional image. The complexity of point cloud data processing is reduced by using the image domain method. Then, to address the issue that the sparse laser point cloud cannot accurately reflect the real environment, the range information of the LiDAR is used as the weight of the adaptive adjustment depth map to set the resolution of the depth map. The depth map in different distance intervals has different resolutions, and the dynamic obstacles can be accurately identified in different distance intervals, which improves the efficiency of the system in identifying dynamic barriers. Finally, the local map is subtracted from the depth map formed by the LiDAR query frame, the dynamic obstacles are removed by the obtained disparity map, and the static points that are mistakenly deleted are recovered by reducing the resolution of the depth map to get a high-precision static map. Experimental results show that the dynamic obstacle removal score of the proposed algorithm is 94.13% in the simulation environment. The dynamic obstacle removal score of the proposed algorithm is 95.22% in the KITTI dataset. The dynamic obstacle removal score of the proposed algorithm is 96.43% in small scenes and 93.30% in large scenes. The algorithm can efficiently remove dynamic obstacles and improve map accuracy and system robustness.