DeepSLAM: A Robust Monocular SLAM System With Unsupervised Deep Learning

In this article, we propose DeepSLAM, a novel unsupervised deep learning based visual simultaneous localization and mapping (SLAM) system. The DeepSLAM training is fully since it only requires stereo imagery instead of annotating ground-truth poses. Its testing takes monocular image sequence as the input. Therefore, SLAM paradigm. consists several essential components, including Mapping-Net, Tracking-Net, Loop-Net, graph optimization unit. Specifically, Mapping-Net an encoder decoder architecture for describing 3-D structure environment, whereas Tracking-Net recurrent convolutional neural network capturing camera motion. Loop-Net pretrained binary classifier detecting loop closures. can simultaneously generate pose estimate, depth map, outlier rejection mask. evaluate its performance on various datasets, find that achieves good in terms estimation accuracy, robust some challenging scenes.