Title

Segmenting and labeling point clouds in VR Bachelorthesis

Type

Research

Year

2019

With the increasing importance of design and construction in existing contexts, methods for "as-built" documentation of buildings are used more and more. Especially since the uprise of BIM, the demand for 3D data increases. Such 3D data, often in the form of point clouds, could come from laserscanning or photogrammetry. To transform these survey data into usable BIM's (Scan to BIM), lots of manual labour is still required. Current programs and methods, especially for the AEC industry, are still at an early stage of development and often very specialized or hard to use. Consequently the motivation arose to look for possibilities to make manual processing of point clouds more intuitive and accessible to a larger audience. Thus, my bachelorthesis deals with the questions of how make the processing of point clouds from the AEC industry more intuitive, which steps are most important and is virtual reality an answer to these questions? The result is a software prototype, which enables the cleaning, segmenting and labeling of point clouds in virtual reality.

For the prototype I focused on three selection algorithms, as seen in the images above. The brush select allows the selection of smaller point groups and is very precise. The radius of the brush and the distance to oneself are controllable. The box select is more focused on larger objects (e.g. construction machines). Since most buildings nowadays are composed of planar elements, the plane select is useful for selecting complete walls or elements within a wall.
Comparing these methods of point selection in VR to existing methods on 2D screens, resulted in the VR methods being slightly faster and more intuitive (for untrained individuals in both softwares). Also the astonishing quality of dense point clouds led to the idea of using survey data for virtual site visits. Merging this with selection and annotating tools to exchange and create semantic data (e.g. Issues, damage maps) to integrate this into BIM processes could be an interesting topic for further research.

Schependomlaan Dataset

github.com/buildingSMART

Semantic3d Dataset

www.semantic3d.net

Unreal Engine 4.22

www.unrealengine.com/en-US/

PointCloudPlugin v.0.5.5

www.pointcloudplugin.com