Real-Time Direct Monocular SLAM With Learning-Based Confidence Estimation

Weiqi Zhang,Aidi Feng,Wangmeng Zuo,Zifei Yan,Gang Xiao

doi:10.1109/access.2019.2928578

Abstract

Direct monocular simultaneous localization and mapping (SLAM) methods, for which the image intensity is used for tracking and mapping instead of sparse feature points, have gained in popularity in recent years. However, feature-based methods usually have more accurate camera localization results than most direct methods, though direct methods can work better in a textureless environment. To tackle the localization issue, we develop a novel real-time large-scale direct SLAM model, namely, GCP-SLAM, by integrating the learning-based confidence estimation into the depth fusion and motion tracking optimization. In GCP-SLAM, a random regression forest is trained off-line with pre-defined confidence measures for learning confidence and detecting the ground control points (GCPs). Then, the confidence value along with the selected GCPs is utilized for depth refinement and camera localization. Our proposed method is shown experimentally more reliable in tracking and relocalization than the previous state-of-the-art direct method when compared with feature-based and RGBD SLAMs.

Highlights

Monocular Simultaneous Localization and Mapping (SLAM) [1]–[3] has received consistent attention in robotics, augmented reality and autonomous cars in recent years
A random regression forest is applied to confidence estimation for small baseline stereo matching in direct monocular SLAM, where the predefined matching cost-based, image-based and depth-aware features are exploited for confidence modeling
Building on the success of confidence estimation based on the random regression forest [21], [22], [25], in this paper, we introduce the learning-based confidence estimation method into monocular SLAM framework for improving the depth estimation and camera localization

Summary

INTRODUCTION

Monocular Simultaneous Localization and Mapping (SLAM) [1]–[3] has received consistent attention in robotics, augmented reality and autonomous cars in recent years. A random regression forest is applied to confidence estimation for small baseline stereo matching in direct monocular SLAM, where the predefined matching cost-based, image-based and depth-aware features are exploited for confidence modeling. By choosing a set of pixels with high confidence as GCPs, an improved camera localization method based on minimizing the photometric and geometric error was provided by increasing the weight term of GCPs in direct image alignment on sim(3). We incorporate the depth refinement and GCPs based camera localization into the LSD-SLAM framework, and achieve accurate and reliable performance in both motion tracking and 3D reconstruction.

RELATED WORKS

EXPERIMENTS

Findings

CONCLUSION