Semantic segmentation of 3D point clouds - Birmingham City photogrammetric data
The video presents a 3D point cloud that shows a stadium in Birmingham, and the surrounding area. We see it both 'unprocessed' (i.e. with its original RGB values) and with annotated semantic labels that show which class each spot of point cloud belongs to (these classes are building, vegetation, road, path, land, car, fence, etc.).
The goal of the video is twofold:
First, the video shows the power of the point cloud to display objects and areas very realistically, with a very high level of detail.
Second, the video shows the power of semantic segmentation, which can very clearly distinguish geometrically close, but semantically very diverse objects.
Additionally, the semantic segmentation of point clouds can extract hard-to-see objects, such as cars, fences, and 'street furniture' in a very clear way.