Modern small-footprint LIDAR systems have the ability to resolve structural details at sub-meter sizes, which make
them ideal for collecting information to use in line-of-sight analysis. Many existing techniques used to map line-of-sight
apply simple surface triangulation to the LIDAR point cloud, but are not well suited to scenes with significant 3D
structure and overlapping objects. Newer voxel-based techniques have the ability to describe scene structure accurately,
but typically suffer from a lack of information if all scene surfaces are not exhaustively sampled by the LIDAR. LIDAR
instrument position is typically discarded after producing the point cloud, but we show how it can be used to identify
areas in voxel maps where insufficient data are available. Using this knowledge of under-sampled areas we demonstrate
construction of an improved line-of-sight map with metrics that indicate where and why errors in the line-of-sight are
likely to occur. During the summer of 2010 an airborne experiment over the RIT campus collected both LIDAR and high
resolution visible imagery. The LIDAR point cloud was sampled at several returns per square meter, and the
accompanying visible imagery is used to provide context and truth information for LIDAR derived products. A realworld
voxel line-of-sight map created from this LIDAR collection is presented along with an analysis of the associated
derived errors.
|