Point cloud filtering is a crucial processing technique for separating ground points from non-ground points and obtaining the most accurate ground point cloud. It serves as the foundation for landslide identification in highway slope point clouds. To address issues such as slow processing speed， low result accuracy， and high error rates encountered by traditional point cloud filtering algorithms in slope scenarios， an airborne slope point cloud filtering algorithm based on multi-plane segmentation and matrix transformation is proposed. This method initially employs a region growing algorithm based on curvature for multi-plane segmentation of the slopes， resulting in multiple sub-point clouds of the slopes. Subsequently， it fits plane models for these sub-point clouds and uses rotation matrices to spatially transform them onto a horizontal plane. Non-ground points are separated by simulating fabric settling with a distance threshold. Finally， the inverse of the rotation matrix is applied for spatial position restoration， yielding the filtered slope point cloud. High-precision point cloud models are obtained using a fine approximation flight route design method for algorithm testing in various slope scenarios. Results are compared with those of other traditional filtering algorithms， demonstrating that the algorithm in this paper outperforms the others with total errors of 7.11%， 4.15%， 1.45%， and 4.41% in all experiments， respectively. Furthermore， the Kappa coefficient values are 0.77， 0.90， 0.96， and 0.90， all of which are the highest among all tested algorithms. The proposed algorithm exhibits high accuracy and applicability， particularly in complex slope scenarios characterized by varying terrains and vegetation cover. It offers a new solution for point cloud filtering in highway slope applications.