Research on Environmentally Robust Monocular Dense SLAM Algorithm Based on Deep Learning

This paper aims at the challenges faced by monocular Simultaneous Localization and Mapping (SLAM) in positioning and 3D reconstruction, and proposes a monocular dense SLAM algorithm with environmental robustness. This algorithm combines the advantages of the classic feature-based SLAM method and the learning-based feature extraction algorithm to improve positioning accuracy and robustness. Meanwhile, it integrates the zero-shot learning monocular depth estimation algorithm, designs a globally consistent dense mapping method, and utilizes environmental structure information for pose recovery to maintain continuous positioning. Experimental results on the EuRoC dataset and complex outdoor campus scenes show that the algorithm has higher positioning accuracy and can effectively handle complex scenarios such as sparse textures and drastic illumination changes, and has application potential in multiple fields such as robotics, AR/VR, etc.