VITAMIN-E: VIsual Tracking and MappINg With Extremely Dense Feature Points

Abstract

In this paper, we propose a novel indirect monocular SLAM algorithm called "VITAMIN-E," which is highly accurate and robust as a result of tracking extremely dense feature points. Typical indirect methods have difficulty in reconstructing dense geometry because of their careful feature point selection for accurate matching. Unlike conventional methods, the proposed method processes an enormous number of feature points by tracking the local extrema of curvature informed by dominant flow estimation. Because this may lead to high computational cost during bundle adjustment, we propose a novel optimization technique, the "subspace Gauss-Newton method", that significantly improves the computational efficiency of bundle adjustment by partially updating the variables. We concurrently generate meshes from the reconstructed points and merge them for an entire 3D model. The experimental results on the SLAM benchmark dataset EuRoC demonstrated that the proposed method outperformed state-of-theart SLAM methods, such as DSO, ORB-SLAM, and LSD-SLAM, both in terms of accuracy and robustness in trajectory estimation. The proposed method simultaneously generated significantly detailed 3D geometry from the dense feature points in real time using only a CPU.

Kimera: From SLAM to spatial perception with 3D dynamic scene graphs

“…Recent work investigates CPU-based approaches in combination with RGB-D sensing (e.g., Wald et al, 2018), PanopticFusion (Narita et al, 2019), and Voxblox++ (Grinvald et al, 2019). A sparser set of contributions addresses other sensing modalities, including monocular cameras (e.g., CNN-SLAM (Tateno et al, 2017), VSO (Lianos et al, 2018), VITAMIN-E (Yokozuka et al, 2019), and XIVO (Dong et al, 2017)) and lidar (Behley et al, 2019; Dubé et al, 2018).…”

SurfaceView: Seamless and Tile-Based Orthomosaics Using Millions of Street-Level Images From Vehicle-Mounted Cameras

“…Approaches in SLAM can generally be classified based on the type of sensors used to sense the environment. LiDAR SLAM uses LiDAR sensors at its core while visual SLAM uses cameras as a main sensor such as monocular-camera [2]- [6], stereo [7]- [9], RGB-D [8], [10], [11], etc. Monocular-based approaches, although widely adopted due to their simple and economical setup, suffer from recovering the metric scale, and that integrating with an IMU (either loosely-coupled or tightly-coupled), referred to with a prefix visual-inertial, offers a main advantage of solving scale ambiguity and providing a more robust navigation.…”

“…AutoStitch 5 , OpenPano 6 , issues with memory management [65], we could only be able to compare our system against Agisoft Photoscan 7 (AS) and Context Capture 8 (CC). We use the same dataset as that used in Pose estimation evaluation, and the computational results are presented in Tab.…”

“…http://matthewalunbrown.com/autostitch/autostitch.html6 https://github.com/ppwwyyxx/OpenPano 7 https://www.agisoft.com 8 https://www.bentley.com/en/products/brands/contextcapture9 Since MGRAPH does not work with oblique images, we used the camera poses computed by our system for image registration and used their approach[63] to create orthomosaics.…”

RGB-D Odometry and SLAM

“…Bundle Adjustment [112], consisting on the joint optimization of a set of camera poses and points, is frequently used to obtain a globally consistent model of the scene [79]. However, there are also several recent VSLAM approaches that alternate the optimization between points and poses, reducing the computational cost with a small impact in the accuracy, given a sufficient number of points [84,94,120,123]. In its most basic form, the map model consists of a set of n points and m RGB-D keyframes.…”