Aiming at the problem that the 3D LiDAR SLAM (Simultaneous Localization and Mapping) algorithm lacks loop-closure detection and are prone to odometry drift when the radar moves too fast, a SLAM method that integrates BA (Bundle Adjustment)/NDT (Normal Distribution Transformation) and 3D LiDAR LOAM is proposed. Firstly, the lidar BA algorithm is improved by minimizing the distance between feature points and edges or planes. Then, the LOAM algorithm is employed, where the front-end estimates coarse poses through frame-to-frame odometry, and the back-end refines the poses using BA. Finally, NDT is utilized to estimate the line and plane feature matrix, performing rotation-invariant transformations and loop closure detection until the latest frame aligns with the initial map, completing the global pose correction. The test is carried out on the public data set KITTI show that the proposed method achieves a root mean square error (RMSE) of 16 cm, reducing the error by about 33% compared to the LOAM algorithm, and the overall running speed is 2.23 times faster than LOAM. To further validate the method in real-world environments, a handheld modeling device is designed, using the pose data from the cartographer algorithm as a reference. The test results demonstrate that the proposed method achieves an RMSE of 20.23 cm, reducing the error by about 75% compared to LOAM, effectively improving positioning accuracy.