基于深度学习的图像深度感知SLAM综述

A review of SLAM based on deep learning image depth perception

关键词:  同步定位与地图构建  深度学习  图像深度估计  里程计  智能定位技术  神经辐射场  三维高斯喷溅

Key words:  Simultaneous localization and mapping (SLAM)  Deep learning  Image depth estimation  Odometer  Intelligent positioning technology  Neural radiance field (NeRF)  3D Gaussian splatting(3DGS)

摘要:

同步定位与地图构建（SLAM）技术在精度和建图方面取得了显著进展，并广泛应用于家用机器人和自动驾驶等领域。随着深度学习和神经网络的快速发展，现阶段的神经网络已经具备从大量数据中学习普适规律的能力，而且还能作为一种新型三维表示方法。基于此，将深度学习与SLAM技术相结合的方法成为了研究热点。概述了SLAM技术与基于深度学习的图像深度感知技术结合的最新进展，对最新的方法进行了总结，并提出了一种可行的框架构建SLAM系统。其中的深度感知技术包括深度估计网络、神经辐射场（NeRF）和三维高斯喷溅（3DGS）技术，详细分析了这3种深度感知技术之间的联系以及它们在SLAM中的潜在应用，为SLAM技术的未来发展提供了一个新的视角，并为进一步的研究提供了参考。

Abstract:

Simultaneous localization and mapping (SLAM) technology has made significant progress in accuracy and mapping, and has been widely used in areas such as home robotics and autonomous driving. With the rapid development of deep learning and neural networks, neural networks at this stage have the ability to learn universal laws from a huge amount of data, and can also be used as a new 3D representation method. Based on this, the method of combining deep learning with SLAM technology has become a research hotspot. Recent advances in combining the SLAM techniques with deep learning-based image depth perception techniques is outlined, the latest approaches are summarized, and a feasible framework for building SLAM systems is proposed, where the depth perception techniques inside include depth estimation networks, neural radiance field (NeRF) and 3D Gaussian splatting (3DGS). The relationship between these three depth perception techniques and their potential applications in SLAM are analysed in detail, offering a new perspective for the future development of SLAM and providing references for future studies.