Improvement Monocular ORB SLAM for Indoor Drone 3D Localization Using Inertial Navigation System

3D localization for indoor drones has challenges due to poor GPS signals caused by building structures. One of popular solution to this problem is to use Monocular ORB SLAM (Oriented FAST and Rotated Brief Simultaneous Localization and Mapping). However, Monocular ORB SLAM is highly dependent on RGB image quality. The accuracy of monocular ORB SLAM will decrease if the input image is dark or blurry, resulting in loss of feature tracking. Missing feature tracking causes position estimation to be inaccurate. To solve this problems, Monocular ORB SLAM will combine with an Inertial Navigation System (INS) to estimate the drone's position when feature tracking fails. INS estimates the drone's speed based on linear acceleration data and estimates the drone's heading based on absolute orientation data. Orientation data is obtained from the drone's built-in Inertial Measurement Unit sensor. In this research, the algorithm for determining pose estimation changes using Monocular ORB SLAM and INS is called ORB+INS Switch. ORB+INS Switch has smaller absolute errors than Monocular ORB SLAM or INS independently. In the feature tracking loss condition, Monocular ORB SLAM has a total position error (TPE) 4.933 m, and ORB+INS Switch has a TPE of 1.008 m. To compare absolute heading error (AHE) values, Monocular ORB SLAM has a more excellent AHE value 16,97°, compared to ORB+INS, which has an AHE value 7,01°.