We present an efficient and accurate point cloud registration method that has been successfully applied to real-world landslide detection tasks. Existing point cloud registration methods often suffer from either low accuracy or high computational complexity, rendering them impractical for real-world applications. In this paper, we propose an efficient and accurate SparsePoint Transformer framework for point cloud registration, which directly applies an attention mechanism to sparse points, significantly enhancing computational efficiency. Additionally, we introduce a cascaded feature aggregation encoder to enrich the contextual details in point clouds, thereby improving registration accuracy. To further adapt our framework for landslide detection, we propose the point cloud registration for landslide detection (PCR4LD) framework, built on the SparsePoint Transformer pipeline. This framework first addresses vegetation interference with two proposed solutions. Subsequently, it employs ICP-based pose refinement for further pose refinement, ensuring accurate landslide detection. Finally, region merging and filtering are applied to identify the landslide-affected regions. Our method not only achieves state-of-the-art registration results on the public KITTI dataset with a significant speedup but also demonstrates outstanding performance in real-world landslide detection tasks, significantly outperforming other methods.