摘要
车载BEV空间配准技术作为智能驾驶车辆的重要手段,一直是自动驾驶领域研究的热点。在BEV空间,如何自适应的提高配准精度,成为自动驾驶BEV空间应用的难点。针对视觉数据深度估计误差会随距离累积、激光数据在低强度场景鲁棒性差和多源配准强依赖于时间同步精度,时序偏差易引发错位等问题,提出环视BEV多层级语义车位加权配准方法,即基于环视BEV拼接图后的语义识别结果,获取车位等语义特征信息,对车位进行多层级跟踪,确定配准初始位姿,基于车载传感器成像精度与距离相关原理,配准平差中引入基于欧式距离的自适应权重计算,从而提高配准精度和鲁棒性。该方法极大提高了配准精度,鲁棒性高,能够较好的基于配准结果进行建图与定位。地下车库环视数据集的建图与定位结果验证了利用本算法进行BEV空间配准的精度与效果。
In-vehicle BEV(Bird′s Eye View)spatial registration technology is an important tool for intelligent driving vehicles and has consistently been a research focus in the field of autonomous driving.In the BEV space,adaptively improving registration accuracy has become a major challenge for BEV applications in autonomous driving.Issues such as the accumulation of depth estimation errors in visual data over distance,the poor robustness of LiDAR data in low-intensity scenarios,and the strong reliance of multi-source registration on time synchronization,where temporal deviations easily lead to misalignment are addressed.A surround-view BEV multi-level semantic parking space weighted registration method is proposed.Specifically,based on the semantic recognition results from the surround-view BEV stitched image,semantic feature information such as parking spaces is obtained,multi-level tracking of these parking spaces is carried out to determine the initial registration pose,and adaptive weights based on Euclidean distance are introduced into the registration adjustment according to the imaging accuracy of in-vehicle sensors and distance-related principles.This enhances registration accuracy and robustness.The method significantly improves registration precision and robustness,enabling effective mapping and localization based on the registration results.The mapping and localization results from an underground parking garage surroundview dataset validate the accuracy and effectiveness of BEV spatial registration using this algorithm.
作者
史兴领
朱添翼
黄刘生
檀杰
赵晓东
SHI Xinling;ZHU Tianyi;HUANG Liusheng;TAN Jie;ZHAO Xiaodong(University of Science and Technology of China,School of Computer Science and Technology,Hefei 230027,China;Guochuang Hefei Intelligent Automobile Technology Co.,Ltd,Perception Fusion Department,Hefei 230088,China)
出处
《光学技术》
北大核心
2026年第1期109-114,共6页
Optical Technique
基金
安徽省科技攻关计划项目(202423k09020037)。
关键词
环视BEV
多层级跟踪
自适应加权
配准
Surround view BEV
multi-level tracking
adaptive weighting
registration
