New Approach for Planar Patch Segmentation Using Airborne Laser Data

Segmentation of LIDAR data has been repeatedly addressed by prior research in the photogrammetric community. The main motivations behind such research are to abstract the huge number of points in a typical LIDAR survey and to relate the LIDAR discrete points to physical objects. The key challenges for an effective and accurate segmentation of LIDAR data are: 1) defining meaningful neighborhoods among irregularly spaced points; 2) deriving useful attributes for the established neighborhoods; and 3) aggregating neighboring points with similar attributes into clusters. In this paper, we present a new LIDAR segmentation technique for planar patches which addresses the above challenges. First, an existing neighborhood definition, which is based on the physical characteristics of the surface, is adopted. The neighborhood of a given point is defined as neighboring points that belong to the same surface. Such a definition has a positive impact on the derived attributes for the defined neighborhoods. Since we are mainly focusing on the segmentation of planar patches, the derived attribute is the magnitude of the surface normal from the origin to the established plane through LIDAR footprints within a given neighborhood. An accumulator array is used to aggregate points with similar attributes into clusters. To avoid ambiguities resulting from the use of the magnitude of the surface normal as the attribute, we adopt an accumulator array that is based on attributes derived from two origins. Since accumulator arrays cluster spatially separate coplanar patches into one group even though they are physically different, the boundaries of derived clusters are used to resolve ambiguities in clustering. It should be noted that the derived boundaries do not represent the actual boundaries of the physical objects being segmented. In other words, these are the boundaries as determined by the LIDAR footprints, and their deviation from the physical boundaries depends on the average point density of the LIDAR data. Therefore, regularization of the extracted boundaries is introduced so that they are as close to the actual boundaries as possible. Experimental results from real data are introduced to highlight the major components of the proposed methodology and these results have proven its feasibility and robustness.