Semantic Map Building Based on Object Detection for Indoor Navigation

Abstract

Building a map of the environment is a prerequisite for mobile robot navigation. In this paper, we present a semantic map building method for indoor navigation of a robot using only the image sequence acquired by a monocular camera installed on the robot. First, a topological map of the environment is created, where each key frame forms a node of the map represented as visual words (VWs). The edges between two adjacent nodes are built from relative poses obtained by performing a novel pose estimation approach, called one-point RANSAC camera pose estimation (ORPE). Then, taking advantage of an improved deformable part model (iDPM) for object detection, the topological map is extended by assigning semantic attributes to the nodes. Extensive experimental evaluations demonstrate the effectiveness of the proposed monocular SLAM method.