LiDAR has been widely used in 3D reconstruction due to its high resolution, wide range and tolerance towards light and weather. To realize accurate and complete environment perception and reconstruction, LiDAR point cloud registration plays a crucial role. This paper utilized an Iterative Closest Point (ICP) algorithm to register the sparse point cloud sensed by LiDAR into a whole indoor environment model. Instead of using a standard ICP algorithm, a point-to-plane ICP is adopted with point cloud selection, point pair matching and rejection. The transformation value between two point cloud data is iteratively calculated and optimized until the defined error metric reaches convergence.