3D Reconstruction and Visual SLAM of Indoor Scenes for Augmented Reality Application

The indoor environments are usually considered as GPS-denied areas, which make the localization of mobile devices a challenging task. The visual SLAM (simultaneous localization and mapping) is a commonly adopted technique for indoor localization and 3D reconstruction. It processes the image information of the scene to derive the location of the camera-equipped device and the 3D structure of the environment. Moreover, many modern mobile devices come with multiple built-in sensors. The integration of multi-sensor information is able to provide more robust localization results compared to the monocular camera based methods. In this paper we implement an indoor augmented reality system based on Project Tango developed by Google. This system is able to construct the 3D environment and reuses the constructed data to relocalize the mobile device. The constructed 3D point cloud can be transformed to the map data, and let the system estimate the navigation path. With the accurate pose estimation provided by Tango, the augmented reality content can keep the original position after a long distance movement.