Transforming 2D Human Data into 3D Model for Augmented Reality Applications

Abstract

Abstract Microsoft's Kinect has been used in many augmented reality applications and most of them for interactive purpose. But since the Kinect is a device with depth sensing technology, it has been used for capturing the figure both 2D and 3D as well. In this research, we would like to create a 3D model or avatar which has the shape just like the actual human. Dealing with 3D avatar requires less computer resource than actual 3D model getting from 3D scanner. The purpose of this work is to transform the 2D measured shape of the human body into the standard 3D avatar. The new Kinect2 for windows with Depth Frame and Skeleton Frame function was used for body measurement. Using T-pose and A-pose position, measurement of the width of parts of the body at each joint (Kinect joint: i.e. Shoulder Center Joint, Spine Joint, Hip Center Joint, Hip Left Joint, Knee Left Joint, Ankle Left Joint, Foot left Joint and etc.) can be done easily. Since both sides of the body were measured (T-pose and A-pose), the circumference at each node can be calculated. In order to transform these data into 3D Model that representing human body, a standard model of 3D avatar was created. The size of the human part (measured at the node) was compared with the standard 3D model. If the measured size is larger than the standard one, then we had to scale the size of the standard model up but if the measured size is smaller, then we have to scale down. To make a perfect 3D model, technique called “Soft Selection” was used. Soft Selection is the technique for gradually scaling each node smoothly to match the second node. If we scale without using this technique, then there will be a bump occurred from one node to other. The 3D avatar model created this way will have the same shape as the human body it represented and can be used in many applications such as virtual fitting room, augmented reality games and etc.