UnLoc: Leveraging Depth Uncertainties for Floorplan Localization

2025-09-14

AI Summary

Monocular depth prediction for planar-map-based indoor localization suffers from two key limitations: (i) lack of explicit uncertainty modeling, and (ii) reliance on scene-specific depth networks, hindering generalization and robustness in dynamic environments. Method: This paper proposes a probabilistic sequence-based camera localization framework that requires no scene-specific training. It explicitly models monocular depth prediction as a probability distribution, tightly integrating uncertainty estimation into a planar-map-constrained sequential optimization pipeline. Crucially, it fully leverages off-the-shelf pre-trained depth models without fine-tuning. Contribution/Results: The approach significantly improves generalization and robustness to environmental changes. On the LaMAR HGE dataset, it achieves a 2.7× improvement in localization recall for long sequences (100 frames) and a 16.7× improvement for short sequences (15 frames), substantially outperforming state-of-the-art methods.

Technology Category

• Intelligent Robots: Localization, Mapping, and Navigation
• Search and Optimization: Learning to Search
• Machine Learning: Calibration & Uncertainty Quantification

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Graph Algorithms and Modeling for the Web: Algorithms and analysis for incomplete, noisy, or partially observed Web-related graphs

Abstract

We propose UnLoc, an efficient data-driven solution for sequential camera localization within floorplans. Floorplan data is readily available, long-term persistent, and robust to changes in visual appearance. We address key limitations of recent methods, such as the lack of uncertainty modeling in depth predictions and the necessity for custom depth networks trained for each environment. We introduce a novel probabilistic model that incorporates uncertainty estimation, modeling depth predictions as explicit probability distributions. By leveraging off-the-shelf pre-trained monocular depth models, we eliminate the need to rely on per-environment-trained depth networks, enhancing generalization to unseen spaces. We evaluate UnLoc on large-scale synthetic and real-world datasets, demonstrating significant improvements over existing methods in terms of accuracy and robustness. Notably, we achieve $2.7$ times higher localization recall on long sequences (100 frames) and $16.7$ times higher on short ones (15 frames) than the state of the art on the challenging LaMAR HGE dataset.

Problem

• Addresses lack of uncertainty modeling in depth predictions
• Eliminates need for per-environment trained depth networks
• Enhances camera localization accuracy in floorplan environments

Innovation

• Probabilistic model with depth uncertainty estimation
• Uses pre-trained monocular depth models
• Eliminates need for per-environment training