Marker-less real-time tracking of texture-less 3D objects from a monocular image

PubDate: November 2017

Teams: The University of Electro-Communications

Writers: Yuki Morikubo；Naoki Hashimoto

PDF: Marker-less real-time tracking of texture-less 3D objects from a monocular image

Abstract

We propose a method for estimating 3D posture of a texture-less object for spatial augmented reality (SAR) by using a monocular image.

In recent years, SAR using image projection technology has been attracting attention. SAR is expected to become one of the next-generation AR technologies, because the image projection technique realizes a virtual object by overwriting the appearance of a real object and achieves physical interaction with them. For overwriting the real object to realize high reality, real-time posture estimation of the projection target is quite important. Because the use of sensors for the posture estimation greatly affects the appearance of the target object, non-contact measurement using a camera is desirable. However, the projection target for SAR is often texture-less in order to achieve better image projection. In addition, if the target contains texture, it is cancelled by the projection. Therefore, it is not possible to use the image feature quantities that are often used in camera-based approaches.

In this study, using the contour information of the projection target, we propose a method that enables the position and the orientation estimation of a texture-less object in real time by using a monocular image.