An ORB-SLAM3 Autonomous Positioning and Orientation Approach using 360-degree Panoramic Video

摘要

In order to solve the problem of insufficient support for 360-degree panoramic data in current mainstream monocular visual SLAM system, this paper proposes a highly universal 360-degree panoramic SLAM method on the basis of ORB-SLAM3 system framework. Firstly, considering that SLAM system does not support panoramic video input effectively, source panoramic video is preprocessed into sequential fisheye images with four perspectives. Secondly, a universal fisheye image calibration method based on three dimensional calibration field is proposed to solve the calibration plate dislocation caused by data splicing of panoramic video shots. Finally, ORB-SLAM3 autonomous positioning and orientation of sequence images from different perspectives is carried out, and its accuracy is analyzed and evaluated. Experimental results of indoor and outdoor datasets show that, SLAM localization and pose determination results from different perspectives segmented from the same panoramic video are relatively consistent with high accuracy. In conclusion, the method proposed in this paper can achieve 360-degree panoramic video autonomous positioning and orientation robustly, with a wide application prospect in various artificial intelligence fields, such as unmanned vehicle driving, mapping of unmanned aerial vehicle exploration in unknown areas, robot autonomous positioning and navigation, etc.