We investigate the quality of structured light 3D scanning on pig skin, muscle and fat. These particular materials are interesting in a number of industrial and medical use-cases, and somewhat challenging because they exhibit subsurface light scattering. Our goal therefor is to quantify the amount of error that various encoding strategies show, and propose an error correcting model, which can bring down the measurement bias considerably. Samples of raw and unprocessed pig tissue were used with the number of sampled surface points Nmeat = 1.2 * 106, Nskin = 4.0 * 106 and Nfat = 2.1 * 106 from 8 different pieces of tissue. With the standard N-step phase shifting method, the bias and RMS errors were found to be 0:45 ± 0:22mm (muscle), 0:51 ± 0:19mm (skin) and 0:14 ± 0:16mm (fat). After applying a linear correction model containing view, light angles and point distances, the bias was almost completely removed on test data, and standard deviations slightly reduced. To our knowledge this is the first quantitative study of the measurement error of structured light with biological tissue.
|Series||DTU Compute-Technical Report-2015|