基于无损卡尔曼滤波的车载双雷达目标位置估计方法

Target location estimation for vehicle dual radar based on unscented Kalman filter

在无人驾驶汽车的研究中,对于传感器探测到的目标进行状态估计是环境感知技术的关键问题之一。本文提出了一种基于无损卡尔曼滤波器的算法,根据所获得的经过标记的雷达数据对目标的位置状态进行预测和更新,从而估计无人驾驶车辆双雷达系统的目标位置。本文中的车载雷达系统由四线激光雷达和毫米波雷达组成,标定后的车辆坐标系为与地面平行的二维坐标系,在此系统和坐标系基础上,在实验场地采集真实雷达数据并进行仿真计算。实验证明,相较于单一传感器,雷达组合模型的测量误差得到有效降低,融合数据精度提高。而相较于目前最常用的扩展卡尔曼滤波算法,车辆行驶方向上的平均位置均方误差从6.15‰下降到4.83‰,与车前轮轴平行的方向上,平均位置均方误差值从4.24‰下降到2.99‰,表明本文算法的目标位置估计更加精确,更接近实际值。此外,在同样的运行环境下,本文算法处理500组雷达数据的平均时间也从5.9ms降低到了2.1ms,证明其有更高的算法效率。

In the research of unmanned vehicle, the state estimation of target detected by sensors is one of the key issues in environmental sensing technology. In this paper, an algorithm based on unscented Kalman filter is proposed to predict and update the position of the target based on the obtained radar data, which is used to estimate the target position of the unmanned vehicle dual radar system. The vehicle radar system in this paper is composed of four lines laser and millimeter wave radar. The calibrated vehicle coordinate system is a two-dimensional coordinate system parallel to the ground. On the basis of the system and coordinate system, the real radar data are collected and simulated in the experimental site. Experiments show that compared with single sensor, the measurement error of radar combination model is effectively reduced, and the accuracy of fusion data is improved. Compared with the most commonly used extended Kalman filtering algorithm, the mean square error of the moving direction of vehicle descends from 6.15 per thousand to 4.83 per thousand. The mean square error value of the average position decreases from 4.24 per thousand to 2.99 per thousand in the direction parallel to the front axle, which indicates that the estimation of the target position of this algorithm is more accurate and closer to the real value. In addition, in the same operating environment, the average time of processing 500 groups of radar data is reduced from 5.9 ms to 2.1 ms, proving that the algorithm has a higher algorithm efficiency.