BundledSLAM: An Accurate Visual SLAM System Using Multiple Cameras

Multi-camera SLAM systems offer a multitude of advantages, primarily stemming from their ability to amalgamate information from a wider field of view. This results in enhanced robustness and improved localization accuracy. In this paper, we present a substantial extension and refinement of a state-of-the-art stereo SLAM system, ORB-SLAM2, with the aim of achieving even higher accuracy. To attain this goal, we initiate the process by mapping measurements from all cameras onto a virtual camera known as BundledFrame. This virtual camera is designed to seamlessly adapt to multi-camera setups, allowing us to effectively fuse data captured from multiple cameras. Furthermore, we leverage extrinsic parameters in bundle adjustment (BA) to achieve precise trajectory estimation. In addition to this, we conduct an extensive analysis of BA’s role within the context of multi-camera scenarios. We explore its impact on tracking, local mapping, and global optimization. Our experimental evaluation involves comprehensive comparisons between ground truth and the state-of-the-art SLAM system. To rigorously assess its performance, we employ EuRoC datasets. The results of our evaluations consistently demonstrate the superior accuracy of our system when compared to existing approaches.