|Timeline:||November 2015 - 31 October 2019|
Point clouds are taking more and more important roles in 3D building modelling and water management. It is necessary to keep point clouds, DTMS and DSMs up-to-date in order to promote better data utilization and decision making. As shown in the figure below, this project aims at detecting changes in the outdated airborne laser point clouds and updating the point clouds as well as DTMs and DSMs using multi-view airborne images and dense image matching (DIM) points. The value of our project is to explore the potential of using dense matching points as a replacement for laser scanning points.
Multi-view airborne images might be effective in ALS data updating. Multi-view images covering a certain area can be obtained with lower costs compared to airborne laser scanning. Highly overlapping airborne images can better cover the complicated urban scene from multiple views compared to traditional nadir-view photogrammetry. 2D Spectral information from imagery and 3D features from point clouds can be incorporated into a more reliable scene classification framework.
Research Problems include: (1) Uncertainty assessment of DIM point cloud quality: The DIM quality needs to be evaluated even if no ground truth is available. Data gaps in DIM data from erroneous matching process might hinder point cloud classification and change detection; (2) Many differences between two point clouds will be visible. It is important to understand which changes are related are related to real changes in the terrain and which of those real changes are relevant for updating; (3) Misclassification: Errors of misclassification in scene classification will be propagated into change detection. It’s a challenge to ensure reliable scene classification.