面向智能网联汽车的BEV感知技术与发展趋势OA

Bird's eye view(BEV)perception has become a fundamental technique for environmental understanding in autonomous driving,due to its unified and interpretable spatial representation.This survey provides a comprehensive re-view of BEV perception technologies tailored for intelligent connected vehicles.It systematically categorizes existing approaches based on sensor modality and deployment configuration,covering vehicle-side,infrastructure-side,and vehicle-infrastructure cooperative scenarios.The review introduces a multi-dimensional framework encompassing vis-ion-only,LiDAR-only,and multimodal fusion methods,and analyzes representative techniques in terms of their design principles and implementation strategies.In addition,this work presents the first consolidated comparison of BEV-re-lated datasets,detailing their sensor setups,task types,and annotation schemes to support standardized benchmarking.Finally,the survey outlines key challenges-such as open-category recognition,unsupervised learning from large-scale data,and robustness under sensor uncertainty-and explores future directions involving end-to-end autonomous driving,embodied intelligence,and large-model-based cooperative BEV perception systems.