Study Of 3D Point-cloud Pre-processing And Grid-model Real-time Multiresolution Rendering

Digital human modeling, which is based on the technology of 3D laser scanning, is of important value in application and broad prospects for development.It is therefore widely used in garment industry, health and rehabilitation, production design, and virtual reality, among others.This paper adopts 3D point-cloud captured by 3D laser human body scanner and high resolution 3D face scanner. Plenty of algorithms are realized such as data resampling, denoising, decimation and real-time LOD rendering of triangle model.The main innovations in the research are as follows:1. A data resamping algorithm based on principal curves and cardinal cubic spline interpolation is adopted. This method effectively solved multi-sensor data merging and denoising.By body sengmentation and data partition, it is also efficiently applied in complicated surface such as 3D human body data.2. An effective and feature-preserving 3D point-cloud smoothing method, based on the multi-featured classification of curvature characteristic, is proposed. This method originated from local plane fitting and bilateral denoising. According to the features, the data set are divided into plane, sub-featured and rich-featured neighborhoods. Considering the different neighborhood types, a bilateral filtering of variable parameters is proposed. And the large calculation of normal direction adjustment is effectively avoided.3. A data reduction algorithm based on local curvature and volume grid is implemented. By improving on the BSP-tree hierarchical clustering algorithm, a priority query adopted method , which enhances the operational efficiency and operability, is presented.4. A sequential progressive meshes based on quadric error metric decimation is constructed. On the basis of PM, a series of data structure such as imformation quering table,triangle queue and vertex mapping sheet are applied to realize the real-time LOD (level of details) rendering which motivated by the interaction of viewing distance, window size, model spinning and shifting.