Point cloud filtering algorithm of water and soil loss in the disturbed area of the power transmission and transformation project based on CSF

The accuracy of laser point cloud data is crucial for soil erosion control in engineering disturbance areas. This study applies the Fabric Simulated Filter (CSF) to filter point clouds collected by a mobile stereo panoramic vision sensor system from disturbed areas of a power transmission project. The CSF algorithm preserves ground points and then performs spatial neighbourhood connected component labelling to segment ground point clouds. Perform elevation normalisation within segments using slope and aspect. The remaining non-grounded points are adaptively filtered through a third-order potential difference. The RMSE between the filtered ground points and the reference data is 0.05 m, the MAE is 0.03 m, and the misclassification rate is less than 1%. This indicates that the filtered data has high accuracy and effectively supports soil erosion monitoring. In the future, further exploration can be conducted on the applicability of this algorithm to complex, disturbed terrain under different seasons and climatic conditions.