RECO: Region-Aware Compensation for Extrinsic Perturbations in Roadside 3D Detection

2026-07-23

AI Summary

This work addresses the performance degradation in roadside 3D object detection caused by minor extrinsic calibration errors—such as camera jitter or drift—that lead to feature misalignment. To mitigate this, the authors propose a region-aware extrinsic compensation framework that learns scene partitioning to separately predict 6-DoF pose offsets for near and far regions. A differentiable sigmoid gating mechanism is introduced to seamlessly fuse geometric information from both regions, enabling continuous BEV sampling and stable optimization. The approach innovatively integrates region-adaptive pose correction, gated fusion, and auxiliary supervision losses based on 2D/3D reprojection alignment, significantly enhancing robustness under calibration uncertainty. Experiments on DAIR-V2X-I and Rope3D demonstrate clear superiority over existing methods, with notably stable and accurate detection performance—especially in yaw and z-axis estimation—under extrinsic perturbations.

Abstract

In intelligent transportation systems, roadside 3D object detection provides wide-area perception crucial for traffic understanding, cooperative early warning, and safe autonomous driving. However, existing methods suffer from high sensitivity to camera extrinsics; even slight deviations (whether manifesting as transient jitter or persistent drift) can be significantly amplified by projective geometry. This cascade results in severe feature misalignment and degraded localization. To mitigate this limitation, we propose RECO, a region-aware extrinsic compensation framework that corrects extrinsics using piecewise 6-DoF pose offsets. RECO predicts a learnable range boundary to partition the scene into near and far regions, estimating region-specific pose corrections. A differentiable sigmoid gate then smoothly blends the two compensated geometries to preserve continuous BEV sampling and facilitate stable optimization. To supervise the refinement of extrinsics, we introduce an auxiliary reprojection loss that compares 2D bounding boxes projected from 3D ground truth against 2D annotations, optimizing it jointly with the standard detection objective. Extensive experiments on the DAIR-V2X-I and Rope3D benchmarks under extrinsic perturbations demonstrate consistent improvements over state-of-the-art baselines across both yaw and $z$-axis deviations. RECO also generalizes from transient perturbations to persistent shifts, maintaining highly competitive performance under strict calibration uncertainty.

Problem

• extrinsic perturbations
• 3D object detection
• camera extrinsics
• feature misalignment
• localization degradation

Innovation

• extrinsic compensation
• region-aware
• 6-DoF pose refinement
• differentiable gating
• reprojection loss