Autonomous flight using RGB-D SLAM with a monocular onboard camera only

Abstract

Real-time Visual SLAM has become increasingly used in robotics for navigation, in particular, RGB-D SLAM has become popular due to its higher pose accuracy compared to Monocular SLAM. However when using an RGB-D sensor the size of the robot and power consumption increases, which may reduce battery life. To deal with that, we propose an algorithm for depth map estimation based on optical flow classification from a static environment obtained from a single camera. We test our algorithm in ORB-SLAM2 for RGB-D cameras, using a monocular camera and compare the performance between our algorithm and an ASUS Xtion RGB-D sensor.