Reliable-loc introduces a resilient LiDAR-based global localization system for wearable mapping devices in complex, GNSS-denied street environments with sparse features and incomplete prior maps.
Key Highlights:
- Dual-Stage Observation Model for MCL: Fuses global and local features into Monte Carlo Localization (MCL), using spectral matching and pose error metrics to refine particle weights in feature-poor scenes.
- Spatial Verification-Based Particle Weighting: Leverages inter-cluster spectral matching scores and pose alignment via SVD to validate local feature correspondences and prevent false convergence.
- Adaptive Localization Mode Switching: Dynamically alternates between Reg (registration-based) and PF (particle filter-based) modes using real-time pose uncertainty from both correspondences and odometry.
- Pose Uncertainty Estimation via Hessian Eigenvalue: Monitors localization reliability using the smallest eigenvalue of the Hessian matrix and standard deviations from covariance estimates.
- Robust in Incomplete or Unmapped Regions: Demonstrates strong recovery and localization performance even in data holes or with unstructured scenes like viaducts and flat walls.
- Real-Time Performance: Achieves ~98 ms per frame on standard hardware, with 5000 particles at initialization and 400 post-convergence, suitable for real-time deployment.
- Quantitative Gains Across Scenarios: Achieves position accuracy of ±2.91 m and yaw accuracy of ±3.74°, outperforming baselines like PF-loc and Reg-loc across 7 diverse urban/campus datasets.
- Failure Detection and Reinitialization: Automatically detects localization failure and reinitializes particles around uncertain poses using Gaussian sampling and MCL exploration.
- Scalable and Generalizable Framework: Validated on a 30 km heterogeneous dataset with helmet-WLS and vehicle-MLS data, covering challenging, dynamic urban scenes.
Paper Resources
- Paper: https://arxiv.org/pdf/2411.07815
- Project: https://zouxianghong.github.io/Reliable-loc/
- Github: https://github.com/zouxianghong/Reliable-loc?tab=readme-ov-file
Related articles from LearnOpenCV:
- MASt3R SLAM: https://learnopencv.com/mast3r-slam-realtime-dense-slam-explained/
- Monocular SLAM in Python: https://learnopencv.com/monocular-slam-in-python/
- LiDAR SLAM: https://learnopencv.com/lidar-slam-with-ros2/
