MDE-VIO: Enhancing Visual-Inertial Odometry Using Learned Depth Priors

2026-02-11

AI Summary

This work addresses the challenge of pose estimation failure in monocular visual-inertial odometry under low-texture environments, where sparse features degrade performance. The authors propose the first method to efficiently integrate dense depth priors generated by a Vision Transformer into the VINS-Mono backend. To enhance geometric robustness, they introduce affine-invariant depth consistency and ordinal constraints, along with a variance-gated mechanism to suppress unstable artifacts. This design preserves real-time capability on edge devices while recovering metric scale consistency. Experimental results on the TartanGround and M3ED datasets demonstrate significant improvements in localization accuracy, achieving up to a 28.3% reduction in absolute trajectory error and effectively mitigating trajectory divergence in complex environments.

Technology Category

• Computer Vision: Motion & Tracking
• Intelligent Robots: State Estimation
• Machine Learning: Calibration & Uncertainty Quantification

Application Category

• Security and Privacy: Large-scale security measurements
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Traditional monocular Visual-Inertial Odometry (VIO) systems struggle in low-texture environments where sparse visual features are insufficient for accurate pose estimation. To address this, dense Monocular Depth Estimation (MDE) has been widely explored as a complementary information source. While recent Vision Transformer (ViT) based complex foundational models offer dense, geometrically consistent depth, their computational demands typically preclude them from real-time edge deployment. Our work bridges this gap by integrating learned depth priors directly into the VINS-Mono optimization backend. We propose a novel framework that enforces affine-invariant depth consistency and pairwise ordinal constraints, explicitly filtering unstable artifacts via variance-based gating. This approach strictly adheres to the computational limits of edge devices while robustly recovering metric scale. Extensive experiments on the TartanGround and M3ED datasets demonstrate that our method prevents divergence in challenging scenarios and delivers significant accuracy gains, reducing Absolute Trajectory Error (ATE) by up to 28.3%. Code will be made available.

Problem

• Visual-Inertial Odometry
• Monocular Depth Estimation
• low-texture environments
• edge deployment
• pose estimation

Innovation

• Visual-Inertial Odometry
• Learned Depth Priors
• Vision Transformer
• Affine-Invariant Depth Consistency
• Edge Deployment