Visual SLAM with Graph-Cut Optimized Multi-Plane Reconstruction

TL;DR

This paper introduces a semantic planar SLAM system that enhances pose estimation and mapping from monocular images by using graph-cut optimization for geometric model fitting and an adaptive parameter strategy, addressing challenges in scene scale and data association.

Contribution

It proposes a novel graph-cut based approach for homography and plane estimation in monocular SLAM, improving robustness over traditional RANSAC methods.

Findings

Enhanced pose accuracy demonstrated on multiple datasets.

Robust geometric model fitting with graph-cut optimization.

Effective adaptive parameter strategy improves system stability.

Abstract

This paper presents a semantic planar SLAM system that improves pose estimation and mapping using cues from an instance planar segmentation network. While the mainstream approaches are using RGB-D sensors, employing a monocular camera with such a system still faces challenges such as robust data association and precise geometric model fitting. In the majority of existing work, geometric model estimation problems such as homography estimation and piece-wise planar reconstruction (PPR) are usually solved by standard (greedy) RANSAC separately and sequentially. However, setting the inlier-outlier threshold is difficult in absence of information about the scene (i.e. the scale). In this work, we revisit these problems and argue that two mentioned geometric models (homographies/3D planes) can be solved by minimizing an energy function that exploits the spatial coherence, i.e. with graph-cut optimization, which also tackles the practical issue when the output of a trained CNN is inaccurate. Moreover, we propose an adaptive parameter setting strategy based on our experiments, and report a comprehensive evaluation on various open-source datasets.