• 2026

This work proposes a novel pipeline for real-time monocular V-SLAM, SLAM-MER, implemented from scratch in C++. SLAM-MER introduces a spatio-temporal querying of 3D points for improved camera localization, and a 3D-2D localization design that leverages recent feed-forward geometry estimators while enabling real-time performance.

• 2025

SAC-GNC is a novel adaptive annealing strategy for Graduated Non-Convexity (GNC) which integrates sample consensus into the classical GNC framework. The addition of sample consensus improved robustness and reduced dependence on predefined parameters. The effectiveness of this approach was demonstrated in 3D registration and pose graph optimization tasks.

• 2024

Novel pixel sampler for Neural Implicit Rendering that leverages the scene's Signed Distance Function (SDF). This approach achieves sharper reconstructions and improved performance in less-observed regions, without requiring additional prior data.

• 2023

BANSAC is a sampling strategy for RANSAC. We derived a dynamic Bayesian network that iteratively updates individual data points' inlier probability. In each iteration, the updated probabilities guide the sampling process. This approach outperforms the best baselines in accuracy, being also more efficient.