摘要
针对路侧单传感器存在感知盲区、多传感器协同感知存在大量冗余信息且位置精度较差、车辆位姿准确值无法在线获取等问题,提出一种基于全局位置精度损失最小化的路侧多传感器目标关联匹配方法。首先,根据各传感器历史数据,以方差最小化为优化目标进行多传感器权重分配;然后,基于级联KM匹配算法处理多传感器存在大量冗余信息的目标级数据,实现当前时刻各传感器数据同一目标关联,判断多源信息与目标的从属关系,并将源于同一目标的数据生成目标集合;最后,根据各传感器权重分配进行加权,生成伪迹点,结合全局位置精度损失最小化方法,以伪迹点坐标系为基准对其他传感器坐标系进行整体校准。试验结果表明:该方法可有效去除目标的多传感器冗余信息,伪迹点速度融合结果的平均绝对误差约0.5 km·h^(-1),伪迹点位置融合结果的平均绝对误差约0.1 m,经过参考系校准后,毫米波雷达已匹配观测点的位置精度损失从5.59 m降至0.32 m,相机已匹配观测点的位置精度损失从1.39 m降至1.32 m。相比于传统JPDA、NNDA关联匹配算法,该方法可降低目标关联匹配融合后的平均位置误差0.134 m,并降低算法的时间复杂度。
A roadside multi-sensor target association matching method based on the minimization of global position precision loss was proposed to address the limitations of roadside sensor perception,such as serious blind spots in roadside single-sensor perception,a large amount of redundant information,poor position accuracy in multi-sensor collaborative sensing,and the unavailability of accurate values of online vehicle poses.First,the variance minimization weight was allocated based on the historical data of each sensor.Second,a cascade Kuhn-munkres matching algorithm was employed to process target-level data with a large amount of redundant information,which grouped the instantaneous sensor data and determined the dependency relationship between the multi-source information data and the target.The target set was generated from the data originating from the same target.Lastly,pseudo-points were generated based on the weight allocation of each sensor,and the pseudo-point coordinate system was adopted as the reference for the overall calibration of other sensor coordinate systems by minimizing the global position precision loss method.The experimental results indicate that redundant information in multi-sensor fusion is effectively removed using the proposed method.The average absolute velocity error of the pseudo-points is approximately 0.5 km·h^(-1),and the average absolute position error of the pseudo-points is approximately 0.1 m.Following the calibration,the position precision loss of the millimeter-wave radar’s matched observation points decreased from 5.59 m to 0.32 m,and that of the camera’s decreased from 1.39 m to 1.32 m.A comparison of the traditional JPDA and NNDA correlation matching algorithms indicated that the proposed method can reduce the average position error of target association matching fusion by 0.134 m,and reduce the time complexity.
作者
辜志强
吉鑫钰
褚端峰
陆丽萍
GU Zhi-qiang;JI Xin-yu;CHU Duan-feng;LU Li-ping(School of Automotive Engineering,Wuhan University of Technology,Wuhan 430070,Hubei,China;Intelligent Transportation Systems Research Center,Wuhan University of Technology,Wuhan 430063,Hubei,China;School of Computer Science and Artificial Intelligence,Wuhan University of Technology,Wuhan 430070,Hubei,China)
出处
《中国公路学报》
EI
CAS
CSCD
北大核心
2022年第3期286-294,共9页
China Journal of Highway and Transport
基金
国家重点研发计划项目(2021YFB2501104)
湖北省重点研发计划项目(2020BAB096,2021BAA181)
重庆市工程研究中心开放课题(21AKC44).
关键词
汽车工程
多传感器融合
坐标系校准
智能网联汽车
数据关联
目标跟踪
automotive engineering
multi-sensor fusion
coordinate system calibration
intelligent and connected vehicle
data association
target tracking
