Comparative Analysis of 3D Human modeling generation using 3D Structured Light Scanning and Machine Learning generation

Abstract

Integration of 3D model production from a single 2D RGB picture using machine learning into mainstream human 3D modelling requires significant evaluation data with an appropriate reference value relative to structured light approach scanning. The purpose of this work is to bridge the gap between the structured light technique and the machine learning algorithm generation method through a comparative analysis based on qualitative criteria such as accuracy and efficiency. The subsequent research was undertaken in two parts. Phase 1 centered on the experimental setup of the data collecting approach utilizing several scanning techniques on the sample model in a controlled setting, whereas phase 2 focuses on the analysis of the subsequent data to determine functional equivalency. The most significant finding of the comparison study is the practicality of the PIFuHD machine learning algorithm with respect to the Artec EVA scanner in terms of efficiency and equivalent precision. Despite the advancements in Machine learning algorithm for generative 3D modeling, major improvements are necessary to achieve functional parity with the structured light scanning approach in terms of accuracy.