Pedestrian Navigation System by Aligning 3D Reconstructions by Visual SLAM with Maps

Abstract

In this research, we propose a navigation system which can estimate a position on a map by aligning a 3D reconstruction result by Visual SLAM on the map. First, the system extracts walls of the building using sight line connecting the cameras and the points. Based on the Manhattan-World hypothesis, building walls are transformed into orthogonal line segments using two orthogonal directions in the building. We generate the floor plan of the building by solving the optimal path problem of the graph with those line segments as edges. By performing the same processing on the map image and comparing the two floor plans, the system aligns the 3D reconstruction result and the map by proposed geometric hashing method.