The ISPRS Benchmark on Indoor Modelling provides a public benchmark dataset to enable performance evaluation and benchmarking of indoor modelling methods. The dataset consists of six point clouds captured by different sensors in indoor environments of various complexities.
If you are interested in participation, instructions on how to download the datasets will be immediately provided after filling this questionnaire
More detailed information about the datasets is provided in the pdf file that can be downloaded here:isprs-archives-XLII-2-W7-367-2017.pdf (1.5 MiB)
This point cloud was captured in one of the buildings of the Technische Universität Braunschweig, Germany, using the Viametris iMS3D system. The data includes both the point cloud and the trajectory of the sensor during the acquisition, and both files contain timestamps. The indoor scene comprises 10 rooms on one floor, which are enclosed by walls with different thicknesses. The scene contains 23 doors, both open and closed, and 7 windows. The building is not furnished, so, the level of clutter, defined as the amount of points belonging to elements that do not constitute the building structure, is low, and it mostly corresponds to the presence of people during the survey.
This point cloud was captured in the same building as TUB1. However, in this case, the sensor Zeb-Revo was used to perform the survey across two floors connected by a staircase. The data consist of the point cloud and the corresponding sensor trajectory, both including timestamps. The first level of the building contains 14 rooms, 8 windows and 23 doors (both open and closed, simple and double), while the second level includes 10 rooms with 13 windows and 28 doors (both open and closed, simple and double). Walls have different thicknesses and ceilings have different heights. As in the previous point cloud, the level of clutter is low.
This point cloud was captured in the office of fire brigade in Delft, The Netherlands. The data acquisition was performed using a Terrestrial Laser Scanner, Leica C10. The indoor scene contains 9 rooms on the same level, 10 doors and 53 windows. The level of clutter in this point cloud is high due to the presence of furniture. The point cloud also contains gaps caused by occlusion due to the static mode of the laser scanning. A complexity of this scene is the presence of curtain walls that can challenge the reconstruction process, especially when they contain windows.
This point cloud represents one room and an entrance hall captured at the University of Vigo, Spain. The survey was performed by a prototype of a backpack-based mobile mapping system (Filgueira et al, 2016), providing a point cloud and the trajectory of the sensor both including timestamps. The scene includes one curtain wall, 20 windows and 7 simple doors (both open and closed), of which two belong to an elevator. The scene also contains stairs to the second floor and several columns with circular cross-section in the middle. The ceiling has different heights and the level of clutter is moderate.
This point cloud was acquired by the sensor Zeb-1 in block B of the engineering building of the University of Melbourne, Australia. The indoor scene comprises 7 rooms on the same floor and 14 doors (both open and closed, single and double), with some walls having different thicknesses. Windows are not visible in this point cloud (they were covered by window blinds). The scene also contains stairs descending to a lower level. The level of clutter is moderate due to the presence of several pieces of furniture.
The Grainger Museum, from Melbourne (Australia), is a Non-Manhattan World building comprising 18 rooms on one floor, 13 doors and 41 windows. The indoor environment is highly cluttered mostly due to the presence of various artefacts. The building was acquired with the Zeb Revo RT system from both indoors and outdoors.