基于ORB-SLAM的智能手机实时同步定位建图与半稠密三位重构实现

The Alization of Real-Time Semi-Dense 3D Reconstruction and Simultaneous Localization and Mapping of Smart Phone Based on Orb-Slam

摘要: 实时定位与环境的三维重构已经成为越来越活跃的研究课题。随着增强现实技术的快速发展，移动设备了解自身位置与周围环境的需求越来越重要。增强现实想要实现追踪、物体识别，实现虚拟数据与现实场景相结合，就必须得到精确而可靠的位姿估计和周围详细的三维场景。相对于其他诸如谷歌眼镜等专用于AR的设备，智能手机作为当前应用范围最广的设备，在智能手机上实现实时定位与制图就变得非常重要。本次研究针对这种情况，基于当前最先进的单目视觉SLAM算法之一ORB-SLAM上实现了智能手机平台上的实时定位与三维重构，并对ORB-SLAM算法进行针对于Android上的优化，使应用可以在近年的智能手机硬件上进行实时定位与三维重构的效果，并且在ORB-SLAM的基础上实现了半稠密的三维重建，可以获得周围环境的轮廓信息而不是地图上稀疏的散点。应用获得到的结果可以用于物体识别与导航以及更有意义并且更加丰富的AR应用。

Abstract: Real-time location and three-dimensional reconstruction of the environment have become an increasingly active research topic. Augmented reality wants to implement tracking, object recognition, data and virtual reality scenarios combining, they must be aware of accurate and reliable pose estimation and around detailed 3D scenes. Compared to others such as Google for AR glasses and other special equipments, smart phone, as the current most widely used equipment, real-time locating and mapping on smartphones become very important. In this case, based on orbslam, one of the most advanced monocular vision slam algorithms, the real-time location and 3D reconstruction on the smartphone platform are realized. The orbslam algorithm is optimized for Android so that applications can be real-time location and the effect of three-dimensional reconstruction, using the smartphone hardware in recent years. And it implements a semi-dense three-dimensional reconstruction on the basis of ORB-SLAM, the contour information may be obtained in surroundings instead of sparse points on the map. The results of the applications got can be used for object recognition and navigation as well as more meaningful and richer AR applications.