渐进三角网加密算法（progressive TIN densification,PTD）在处理激光雷达点云数据时选择参数单一，大面积综合地形处理时自适性较差。基于此问题提出了一种多源数据融合的点云滤波方法，该方法首先对可见光遥感影像进行分类及后处理，再进一步融合激光雷达点云数据，通过PTD滤波法实现区域自适应滤波分离地面点。结果表明：基于面向对象的SVM分类方法相比于随机森林分类法在处理可见光遥感影像时效果显著，其平均总体分类精度最高为88.37%，平均Kappa系数最高为0.86；基于融合数据的PTD滤波法较单一PTD滤波法可实现区域自适应滤波，在处理复杂山区地形时可自动选择合理的参数进行点云滤波提取，极大地提高了地面点滤波精度。

The progressive TIN densification(PTD) algorithm faces limitations in parameter selection for laser point cloud data, particularly in comprehensive terrain processing. According to this problem, we proposed a point cloud filtering method for multi-source data fusion. We classified and post-processed optical remote sensing images at first. Then, we integrated laser point cloud data, and separated the ground points by regional adaptive filtering through the PTD filtering method. The results show that object-oriented SVM classification outperforms random forest classification for optical remote sensing images, with the highest average overall accuracy reaching 88.37% and the highest average Kappa coefficient reaching 0.86. The PTD filtering method based on fusion data demonstrates superior adaptability compared to singular PTD filtering, automatically selecting optimal parameters for point cloud filtering in complex mountainous terrains, significantly enhancing ground point filtering accuracy.