To address the challenges faced by conventional lidar systems in accurately filtering out water spray and dust, highly reflective expansion crosstalk noise, and the suboptimal removal of large-scale outliers, a laser point cloud denoising algorithm based on ground fitting and Density-Based Spatial Clustering of Applications with Noise (DBSCAN) is proposed. Initial noise location information is obtained through ground point cloud segmentation, an enhanced DBSCAN algorithm, and a three-dimensional bounding box estimation method. Noise is then marked and filtered out based on its geometric and physical characteristics. The experimental results demonstrate that in dust and water spray scenarios, the noise-reduced Chamfer Distance (CD) decreases from 0.037 4 m to 0.019 8 m, while the Point-to-Mesh (P2M) distance reduces from 0.158 m to 0.088 m. The overall geometric deviation of the point cloud is significantly diminished, with improved conformity to the true surface. For processing complex point cloud data, the optimised DBSCAN clustering achieves a computational time of merely 6 ms compared to the conventional DBSCAN algorithm, representing a 77.8% improvement in computational efficiency. This effectively resolves the suboptimal outlier filtering performance of traditional laser point cloud denoising algorithms at large scales.