3D Human Reconstruction: A Comparison Between Images-Based and Point Cloud-Based Method

Abstract

Nowadays, Human Reconstruction is an important part of 3D Reconstruction because it enables us to acquire human information. In reality, Human Reconstruction is applied more and more widely in various domains of people's life such as healthcare, education, sport, entertainment, fashion, anthropometry…especially in the Covid-19 pandemic situation. Technically, the digitalization of human shape and pose is being implemented by two main solutions. The first solution aims to re-create 3D shape and pose of people by using a set of 2D images from different view angles. The second method focuses on handling with the point cloud which is acquired from a set of depth cameras. Both of them have their own advantages and disadvantages. The current trend of 3D Human Reconstruction is to try recreating human shape and pose not only from various available sources (images, point cloud) but also from natural sources (surveillance camera). However, the accuracy of current solutions is significantly sensitive to the working environment. Therefore, this work gives an attempt to develop two human reconstruction methods and evaluate their performance in various conditions as synthetic, laboratory, and real-world environments. The results show a comparison of human reconstruction's accuracy due to the working environment and method of approach. In the synthetic environment, the accuracy is highest for both two methods because of the perfection of input. The accuracy of these methods is lower in the remaining environment. Besides, the result also proves that point cloud-based method is more sensitive than image-based method due to the limitation of hardware devices.