Fine-Grained Point Cloud Semantic Segmentation of Complex Railway Bridge Scenes from UAVs Using Improved DGCNN

Figure 5

Denoising results under different methods. (a) Point cloud before denoising; (b) direct denoising results using the original algorithm; (c) applying window sliding denoising trick based on the original algorithm; and (d) add Gaussian noise to the result of (c) for secondary denoising to enhance the point cloud contours. The example in Figure 5 shows the result of adding 30% Gaussian noise and then secondary denoising.

(a)

(b)

(c)

(d)