ORB-ODOM: Stereo and odometer sensor fusion for simultaneous localization and mapping

Abstract

ORB-SLAM2 is one of the better-known open source SLAM implementations available. However, the dependence of visual features causes it to fail in featureless environments. With the present work, we propose a new technique to improve visual odometry results given by ORB-SLAM2 using a tightly Sensor Fusion approach to integrate camera and odometer data. In this work, we use odometer readings to improve the tracking results by adding graph constraints between frames and introduce a new method for preventing the tracking loss. We test our method using three different datasets, and show an improvement in the estimated trajectory, allowing a continuous tracking without losses.