Multiview range-image registration for forested scenes using explicitly-matched tie points estimated from natural surfaces

Range images collected from terrestrial laser scanners hold promise for characterizing forest canopies and facilitating research into linkages between forest structure and function. Applications that aim to make plot-level or canopy-level assessments may ultimately require combining range imagery from multiple viewpoints, a procedure of image co-registration that poses numerous challenges. We developed a method for simultaneously registering multiple range images using explicitly-matched, computer-estimated tie points from natural ground and tree-stem surfaces in two stages. Stage I involved registration using tie points estimated from ground-surfaces, while stage II involved tie points estimated from tree-stem centers. Eight range images were simultaneously aligned based on more than 3000 tie point pairs from ground surfaces, and more than 4000 tie point pairs from tree stems. Tie points were widely dispersed throughout the scanned volume, spanning a horizontal area of 750 m2 and heights up to 22 m above ground. Results showed accurate image alignments at heights well above ground, with an average post-registration error, for tree-stem-based tie point pairs, of 0.16 cm. When linked to automated algorithms for identification and extraction of tree stems from point clouds, the method developed here should be useful for delivering an array of products related to the detailed assessment of forest canopies and stand structure.