Abstract:Recently, generating dense maps in real-time has become a hot research topic in the mobile robotics community, since dense maps can provide more informative and continuous features compared with sparse maps. Implicit depth representation (e.g., the depth code) derived from deep neural networks has been employed in the visual-only or visual-inertial simultaneous localization and mapping (SLAM) systems, which achieve promising performances on both camera motion and local dense geometry estimations from monocular images. However, the existing visual-inertial SLAM systems combined with depth codes are either built on a filter-based SLAM framework, which can only update poses and maps in a relatively small local time window, or based on a loosely-coupled framework, while the prior geometric constraints from the depth estimation network have not been employed for boosting the state estimation. To well address these drawbacks, we propose DiT-SLAM, a novel real-time Dense visual-inertial SLAM with implicit depth representation and Tightly-coupled graph optimization. Most importantly, the poses, sparse maps, and low-dimensional depth codes are optimized with the tightly-coupled graph by considering the visual, inertial, and depth residuals simultaneously. Meanwhile, we propose a light-weight monocular depth estimation and completion network, which is combined with attention mechanisms and the conditional variational auto-encoder (CVAE) to predict the uncertainty-aware dense depth maps from more low-dimensional codes. Furthermore, a robust point sampling strategy introducing the spatial distribution of 2D feature points is also proposed to provide geometric constraints in the tightly-coupled optimization, especially for textureless or featureless cases in indoor environments. We evaluate our system on open benchmarks. The proposed methods achieve better performances on both the dense depth estimation and the trajectory estimation compared to the baseline and other systems.

Omnidirectional Dense SLAM for Back-to-back Fisheye Cameras

OmniSLAM: Omnidirectional Localization and Dense Mapping for Wide-baseline Multi-camera Systems.

Object Detection and Localization in 3D Environment by Fusing Raw Fisheye Image and Attitude Data

Real-time Omnidirectional Visual SLAM with Semi-Dense Mapping

Stereo Vision Based SLAM Using Rao-Blackwellised Particle Filter

Omnidirectional Depth Estimation With Hierarchical Deep Network for Multi-Fisheye Navigation Systems

Omnidirectional DSO: Direct Sparse Odometry With Fisheye Cameras

Real-Time Dense Mapping for Self-driving Vehicles using Fisheye Cameras

Spatio-Temporal Depth Recovery of Dynamic Scenes with Multiple Handheld Cameras

View-based Underwater SLAM Using a Stereo Camera

Direct Depth SLAM: Sparse Geometric Feature Enhanced Direct Depth SLAM System for Low-Texture Environments

DiT-SLAM: Real-Time Dense Visual-Inertial SLAM with Implicit Depth Representation and Tightly-Coupled Graph Optimization

A photogrammetric approach for real‐time visual SLAM applied to an omnidirectional system

Omnidirectional Multicamera Video Stitching Using Depth Maps

BAMF-SLAM: Bundle Adjusted Multi-Fisheye Visual-Inertial SLAM Using Recurrent Field Transforms

A*SLAM: A Dual Fisheye Stereo Edge SLAM

Real-time Multi-view Omnidirectional Depth Estimation System for Robots and Autonomous Driving on Real Scenes

360ORB-SLAM: A Visual SLAM System for Panoramic Images with Depth Completion Network

MODE: Multi-view Omnidirectional Depth Estimation with 360-degree Cameras

MGS-SLAM: Monocular Sparse Tracking and Gaussian Mapping with Depth Smooth Regularization

FisheyeDepth: A Real Scale Self-Supervised Depth Estimation Model for Fisheye Camera