UW Deep SLAM-CNN Assisted Underwater SLAM

Abstract Underwater simultaneous localization and mapping (SLAM) poses significant challenges for modern visual SLAM systems. The integration of deep learning networks within computer vision offers promising potential addressing these difficulties. Our research draws inspiration from approaches applied to interest point detection matching, single image depth prediction underwater enhancement. In response, we propose 3D-Net, a learning-assisted network designed tackle three tasks simultaneously. consists branches, each serving distinct purpose: detection, descriptor generation, prediction. detector generator can effectively serve as front end classical system. predicted information is akin virtual camera, opening up possibilities various applications. We provide quantitative qualitative evaluations illustrate some uses. was trained in several steps, using in-air datasets followed by generated datasets. Further, the integrated into feature-based SALM systems ORBSLAM2 ORBSSLAM3, providing comprehensive assessment its effectiveness navigation.