In LiDAR-based SLAM algorithms, the process of point cloud registration stands as a pivotal step. When the actual sensor is used to collect point cloud data, the quantity of points within the collected point cloud is frequently immense. Traditional point cloud registration algorithms cannot effectively and quickly handle the registration of dense point clouds. This paper presents a Downsample-based Improved Dense Point Cloud Registration Framework. On the basis of ensuring the registration accuracy, the registration of tens of millions of point clouds can be quickly realized, which saves a lot of time for the entire mapping process. After experimental verification, the algorithm can realize the registration of tens of millions of point clouds within 2 min, which provides a solution for the registration of dense point clouds.