A PbMap is a highly compact representation of the scene based on a planar model of it. This map representation is proposed to avoid the high memory requirements and processing cost of traditional point cloud representations, which use has raised considerably with the appearance of low cost RGB-D (Kinect like) sensors. A PbMap compresses the point cloud into a set of planar patches, neglecting the non planar data. In this way, it offers an enormous data compression at the cost of losing the non-planar details, but we argue that such details have little importance for some applications, as for building lifelong maps, since only the large planes belonging to the scene structure are persistent over time, while the non-planar, generally small objects are more likely to move or disapear from the scene.
- Fernández-Moral, E. and Mayol-Cuevas, W. and Arévalo V. and González-Jiménez, J. Fast place recognition with plane-based maps. IEEE International Conference on Robotics and Automation (ICRA), 2013. (PDF)
This method relies on an interpretation tree to efficiently match sets of neighboring planes. Such interpretation tree applies geometric and radiometric restrictions in the form of both unary and binary constraints. The unary constraints check the individual correspondence of two planes by comparing directly their features (e.g. size, color), while the binary constraints validate if two pairs of connected planes present the same geometric relationship (e.g. perpendicularity).
Further details regarding matching with geometric constraints can be found in:
- Grimson, W.E.L.: Object Recognition by Computer – The role of Geometric Constraints. MIT Press, Cambridge, MA. 2003.
Two example programs can be found in the following directories:
The PbMap user guide can be downloaded here.