DF-Calib: Targetless LiDAR-Camera Calibration via Depth Flow

2025-04-02

AI Summary

To address extrinsic parameter drift between LiDAR and camera induced by mechanical vibrations, this paper proposes a target-free, end-to-end online joint calibration method. The calibration problem is reformulated as a cross-modal dense depth flow estimation task: a shared encoder aligns image-predicted depth maps with sparsely projected LiDAR depth. We introduce a novel depth-flow-driven calibration paradigm, incorporating a reliability map guidance mechanism and a perception-weighted sparse flow loss to significantly improve alignment accuracy in critical regions and enhance robustness against vibration-induced disturbances. The method operates in real time without requiring external calibration targets or prior motion constraints. Evaluated on the KITTI dataset, it achieves state-of-the-art performance with 0.635 cm translational error and 0.045° rotational error, demonstrating strong generalization across diverse driving scenarios.

Technology Category

• Computer Vision: Multi-modal Vision
• Machine Learning: Calibration & Uncertainty Quantification
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Systems and Infrastructure for Web, Mobile and WoT: Experiences and lessons learnt from Web-based algorithms and system deployments
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Precise LiDAR-camera calibration is crucial for integrating these two sensors into robotic systems to achieve robust perception. In applications like autonomous driving, online targetless calibration enables a prompt sensor misalignment correction from mechanical vibrations without extra targets. However, existing methods exhibit limitations in effectively extracting consistent features from LiDAR and camera data and fail to prioritize salient regions, compromising cross-modal alignment robustness. To address these issues, we propose DF-Calib, a LiDAR-camera calibration method that reformulates calibration as an intra-modality depth flow estimation problem. DF-Calib estimates a dense depth map from the camera image and completes the sparse LiDAR projected depth map, using a shared feature encoder to extract consistent depth-to-depth features, effectively bridging the 2D-3D cross-modal gap. Additionally, we introduce a reliability map to prioritize valid pixels and propose a perceptually weighted sparse flow loss to enhance depth flow estimation. Experimental results across multiple datasets validate its accuracy and generalization,with DF-Calib achieving a mean translation error of 0.635cm and rotation error of 0.045 degrees on the KITTI dataset.

Problem

• Precise LiDAR-camera calibration for robotic perception
• Targetless calibration for misalignment correction in autonomous driving
• Improving cross-modal alignment robustness via depth flow estimation

Innovation

• Targetless LiDAR-camera calibration via depth flow
• Shared feature encoder for consistent depth features
• Perceptually weighted sparse flow loss enhancement