Photogrammetric point clouds: quality assessment, filtering, and change detection

Abstract

Chapter 1 -Introduction (iii) Irrelevant changes: These changes happen in reality and also present in the two datasets, but are not the changes we are interested in. For example, leaves grow and fall in different seasons, vehicles and pedestrians moving, water surface fluctuation, a newly-built scaffold, a new container in ports, etc.

Three Dimensional Change Detection Using Point Clouds: A Review

“…Although the handcrafted algorithm can achieve relatively balanced results on the overall and per-class accuracy and mean intersection over union (mIoU), it was obvious that learning-based methods achieved overwhelming performance. In [124], the authors proposed a method to detect building changes between LiDAR (Light Detection And Ranging) and photogrammetric point clouds. With consideration of the fact that semantic segmentation and CD are correlated, they suggested the Siam-PointNet++ model to combine the two tasks in one framework (see Figure 11).…”

“…Only feature vectors extracted at the same centroids can be compared. The authors point out that even if there is no DIM point in the conjugate ball of a LiDAR point, a pseudo feature map is calculated at the same centroid to "inform" the model that the neighborhood of the ball in the DIM data is empty [124]. Experiments conducted in a study area in Rotterdam, Netherlands, indicated that the network was effective in learning multi-task features.…”