Spatiotemporal Calibration, Fusion Strategies, and Future Sensing Modalities for Robust Roadside 3-D Object Detection: A Review

Abstract

Infrastructure-based perception systems are pivotal to the advancement of vehicle-to-everything (V2X) cooperative autonomous driving, providing a comprehensive macroscopic perspective that mitigates the blind spots and limited sensing horizon inherent to onboard sensors. LiDAR and camera fusion has become the mainstream solution for reliable traffic monitoring. By leveraging complementary geometric and semantic information, this approach significantly enhances 3-D object detection performance. However, unlike vehicle-side perception, roadside systems confront unique challenges regarding long-term robustness, specifically spatiotemporal misalignment caused by the mechanical instability of sensor poles and feature degradation under adverse weather conditions. This article provides a comprehensive survey of technologies aimed at achieving robust roadside 3-D object detection. First, we systematically review existing infrastructure-side datasets, revealing a critical scarcity of data covering adverse weather and multispectral modalities compared to vehicle-side benchmarks. Second, we provide an in-depth analysis of spatiotemporal calibration technologies, identifying trajectory-based online calibration as the most viable solution for maintenance-free operation against dynamic sensor drift. Third, we categorize state-of-the-art multimodal fusion architectures, elaborating on the paradigm shift from feature-level projection to unified bird’s-eye-view (BEV) representations and Transformer-based interactions. Furthermore, we discuss infrastructure-specific optimization strategies, including height-aware view transformation, static background suppression, and scenario generalization. Crucially, this survey identifies the physical sensing gap of current visible-light systems in extreme lighting conditions and proposes the integration of thermal imaging as a frontier modality for achieving true all-weather robustness. Finally, open challenges such as edge–cloud synergy and collaborative perception are discussed.