Toward Robust Collaborative Perception under Adverse Weather Conditions Via Dual-Branch Network
Yuquan Yang, Hui Zhang, ZiYin Zhang, Wenyu Lu, Xiaohua Xu
AI summary
Problem
Collaborative LiDAR perception degrades under adverse weather due to feature attenuation and noise contamination, but existing fusion methods treat these as a single pattern, leading to suboptimal performance.
Approach
The method decomposes weather-induced LiDAR degradation into attenuation and contamination, processing them with a specialized dual-branch network (enhancement via radar-guided attention and suppression via consistency checks) integrated by a dynamic decision network.
Key results
- Decomposes weather-induced LiDAR degradation into attenuation and contamination patterns
- Introduces a dynamic decision network for adaptive, spatially-aware branch integration
- Achieves +3.65% mAP@0.7 improvement over baselines under fog conditions
- Achieves +10.80% mAP@0.7 improvement over baselines under snow conditions
Why it matters
Enables reliable multi-agent autonomous driving perception in harsh weather, addressing a critical safety gap for real-world deployment.
Abstract
Recent advances in collaborative perception sys- tems have led to significant improvements in 3D object detection performance. While widely deployed LiDAR and camera sys- tems often experience performance degradation under adverse weather conditions, weather-robust 4D radar offers a promising alternative to address this challenge. However, effectively fusing 4D radar measurements with degraded LiDAR data remains a critical challenge. In this work, we decompose the weather- induced degradation in LiDAR perception into feature attenua- tion requiring enhancement and feature contamination requir- ing suppression, based on the underlying physical interactions. Building upon this decomposition, we propose a dual-branch network to handle each degradation pattern in a specialized manner. One branch focuses on enhancement based on spatial and channel attention, guided by 4D radar cues. The other branch focuses on suppression based on intra-modal structural consistency and cross-modal consistency. To achieve adaptive branch integration, we propose a dynamic decision network to generate a decision weight map for each branch and capture the complex interaction between branches. To validate the effectiveness of our method, we conduct extensive experiments on V2X-R, the only publicly available collaborative LiDAR- 4D radar dataset. Extensive experimental results demonstrate that our method achieves improvements of 3.65% and 10.80% in mAP@0.7 under fog and snow conditions, respectively, outperforming previous state-of-the-art approaches.