Map Projections

Coordinate Systems and Indoor Cartography

5107.4 - Semi-automated alignment of building models to 3D point cloud data

Wednesday, July 5
9:30 AM - 9:50 AM
Location: Maryland B

Over the past two decades, automated navigation systems and location-based services (LBS) have become indispensable features of everyday life. Navigation systems have revolutionized the way people move about their environments by providing real-time wayfinding instructions for all modes of transport to include walking, automobiles, ships, and planes. Location-based services have also permeated society's increasingly wired lifestyles, ranging from openly-consented services such as the geotagging of photos and social media posts to less-noticed services such as the crowdsourcing of traffic data for online mapping. The geographic scope of these services continues to grow with each passing year, yet one environment remains largely in the dark: indoor spaces. Overhead obstructions prevent global navigation satellite systems (GNSS) receivers from properly working indoors, and indoor spaces provide a much more challenging environment for mapping due to their darker, more dynamic, and more heterogenous nature compared to outdoor spaces.

One major challenge for indoor mapping involves the integration of indoor and outdoor coordinate reference systems (CRS). Indoor maps nearly always use local engineering CRSs based on a Cartesian coordinate system referenced to local engineering datums, e.g., local monuments, survey markers, or other fixed points. However, outdoor maps can exist in many different CRSs using either angular geographic units of measurement or linear projected units of measurement, but they nearly always use some part of Earth (e.g., ellipsoid surface, geoid surface, geocenter, etc.) as the datum. The large number of these Earth-based, or geodetic, CRSs complicates the task of integrating indoor and outdoor maps. Fortunately, pre-defined transformations exist between most of these geodetic CRSs.

Our proposed method of integrating indoor and outdoor CRSs assumes the existence of an accurate geometric building model in local coordinates and a point cloud of the building exterior in projected geodetic coordinates. The building model can be custom-built or derived from existing models, such as used in building information modeling (BIM) or CityGML, and the point cloud can come from remote sensing, e.g., LiDAR or photogrammetry. We assume sufficient accuracies exist in both the building model and point cloud to enable rigid body, affine transformations. The goal involves aligning the building model with the point cloud to determine the local-to-geodetic transformation matrices; further transformations can be accomplished using a variety of transformation tools.

Alignment of the model to the point cloud involves three general steps. First, a global alignment is performed by manually establishing correspondences between three tie points in each of the model and the point cloud. This brings the building model into rough alignment with the point cloud. Second, the point cloud is filtered using an expanded convex hull of the building model to ensure only relevant, local points are used in the third step. Finally, an automated iterative process is used to continually refine the building-to-point cloud alignment using a variation of the iterative closest point (ICP) algorithm, originally proposed by Besl and McKay (1992). This variation of ICP minimizes the residuals of predicted errors by rotating and translating the model and progressively prunes the point cloud of outliers as the fit improves.

We performed three studies to validate our approach: a single building, multiple buildings with a single point cloud, and an emergency response situation with a capsized cruise ship. In each study, our method effectively aligned the building models with the geodetic point clouds, with the iterative alignment providing significant improvements over the initial three-point transformation. Future improvements to our proposed method involve refining closest point calculations, refining the criteria for ending iterative rotations, and improving the pruning process.

Jorge Chen

Graduate Student
Department of Geography, University of California, Santa Barbara

Jorge Chen is a doctoral candidate in the Geography program at the University of California, Santa Barbara. He holds a B.S. in Civil Engineering, an M.S. in Civil Engineering, and an M.A. in Geography. His research interests include cartography, geodesy, surveying, indoor mapping and modeling, and gamification using real-world geographic information.


Send Email for Jorge Chen

Keith Clarke

Department of Geography, University of California, Santa Barbara

Dr. Keith C. Clarke is a research cartographer and professor with the M.A. and Ph.D. from the University of Michigan in Analytical Cartography. He is the former North American Editor of the International Journal of Geographical Information Systems and has authored about 250 book chapters, journal articles, and papers in the fields of cartography, remote sensing, and geographic information systems. He has also served on numerous National Research Council studies and the National Geographic Society's Committee on Research and Exploration. Awards include the USGS John Wesley Powell Award (2005) and the Cartography and Geographic Information Society's Distinguished Career Award (2014).


Send Email for Keith Clarke

Miljenko Lapaine

University of Zagreb, Faculty of Geodesy

Miljenko Lapaine graduated from the Faculty of Science, University of Zagreb, in the field of Theoretical Mathematics. He obtained his PhD from the Faculty of Geodesy, University of Zagreb with a dissertation entitled Mapping in the Theory of Map Projections. He has been a full professor since 2003. He has published more than 900 papers, several textbooks and monographs. Prof. Lapaine is the Chairman of the ICA Commission on Map Projections, a founder and President of the Croatian Cartographic Society and the Executive editor of the Cartography and Geoinformation journal.


Send Email for Miljenko Lapaine


5107.4 - Semi-automated alignment of building models to 3D point cloud data

Attendees who have favorited this

Please enter your access key

The asset you are trying to access is locked. Please enter your access key to unlock.

Send Email for Semi-automated alignment of building models to 3D point cloud data