With the expansion of production scale and continuous cost reduction of advanced sensor LiDAR, point cloud registration technology has found widespread application in fields such as autonomous driving and building surveying. In view of the problems with NDT (normal distributions transform) algorithm in registering large-scale point cloud data such as strong dependence on the initial position and low registration efficiency, a NDT point cloud registration algorithm based on PCA (principal component analysis) is proposed. First, the PCA algorithm is used for initial alignment of point clouds, so as to ensure that two sets of point clouds have good initial position. Subsequently, the point cloud is accurately registered utilizing the NDT algorithm based on this foundation. The algorithm’s effectiveness is shown through both qualitative and quantitative assessments. The results from the experiments demonstrate that the algorithm significantly improves the efficiency of point cloud registration, and the precision has also witnessed an increase.