Noise-robust transparent visualization of large-scale point clouds acquired by laser scanning
Tomomasa Uchida, Kyoko Hasegawa, Liang Li, Motoaki Adachi, Hiroshi Yamaguchi, Fadjar I. Thufail, Sugeng Riyanto, A. Okamoto, Satoshi Tanaka
Abstract
We propose a high-quality transparent visualization method suitable for large-scale laser-scanned point clouds. We call the method “stochastic point-based rendering (SPBR),” which is based on a novel stochastic algorithm. SPBR enables us to clearly observe the deep interior of laser-scanned 3D objects with the correct feeling of depth. The high quality of SPBR originates from the effect of “stochastic noise transparentization,” which is an effect to make the measurement noise transparent and invisible in the created images. We mathematically prove that this effect also makes the created transparent images coincide with the results of the conventional methods based on the alpha blending, which is time-consuming and impractical for large-scale laser-scanned point clouds. We also demonstrate the effectiveness of SPBR by applying it to modern buildings, cultural heritage objects, forests, and a factory. For all of the cases, the method works quite well, realizing clear and correct 3D see-through imaging of the laser-scanned objects.