多传感器信息融合的车辆轨迹感知设计与实现

Design and implementation of vehicle trajectory perception with multi⁃sensor information fusion

针对全息路口需求,设计了一种雷达、GPS及视觉的多维信息融合雷视一体机。基于深度学习技术实现对机动车的实时检测。在此基础上,通过雷达与视频及雷达与GPS坐标系标定建立目标关联关系,利用融合规则获取融合数据点。为了解决融合数据点形成的轨迹线不平滑的问题,提出加权GS-LSPIA算法,将每个融合数据点赋予初始权重,选取拟合曲线的控制点,通过迭代的方式调整控制点,生成一系列曲线,曲线的极限是融合数据点的最小二乘拟合的结果。实验结果表明,雷视一体机设备可感知车辆轨迹,且加权GS-LSPIA算法使得车辆运行轨迹更平滑。

A radar⁃video integrated machine based on the fusion of radar,GPS and visual information is designed to meet the requirements of holographic traffic intersections.A real⁃time detection and tracking scheme for motor vehicle targets is designed based on deep learning technology.On this basis,the target correlation relationship is established through the calibration of radar and video and radar and GPS,and fusion rules are used to obtain fusion data points.In order to solve the problem that the trajectory formed by the fusion data points is not smooth,a data⁃weighted GS-LSPIA algorithm is proposed,which assigns an initial weight to each fusion data point,and selects the control points of the fitted curve.In an iterative manner,the control points are adjusted to generate a series of curves,the limits of which are the results of least squares fitting of the fused data points.The experimental results show that the radar⁃video integrated machine can sense the vehicle trajectory,and the data⁃weighted GS-LSPIA algorithm makes the vehicle running trajectory line smoother.