ADAS系统视觉与毫米波雷达分布式抗差卡尔曼滤波融合算法

Vision and Radars Fusion Algorithm Based on Distributed Robust Kalman Filters in ADAS

自动驾驶车辆常常利用多传感器对周围目标进行检测和跟踪,但受限于传感器多源异构特性和复杂多变的驾驶环境,准确的多目标检测和跟踪仍是实现自动驾驶的一大困难和挑战。本文面向高级驾驶辅助系统(ADAS)的多目标检测与跟踪任务,采用了基于1个视觉传感器和5个毫米波雷达(1V5R)的传感器配置方案,且设计了基于分布式抗差卡尔曼滤波器的多传感器信息融合算法以实现对周围目标的准确感知。首先,针对不同传感器数据特征,采用不同的线性卡尔曼滤波器和扩展卡尔曼滤波器进行数据融合,并基于分布式卡尔曼滤波建立了1V5R多传感器信息融合框架。其次,为降低传感器动态误差对于融合精度的影响,在卡尔曼加权观测融合的基础上,引入抗差估计方法,实现了对传感器动态误差的实时估计和修正。最后,通过离线仿真和实车道路试验对所提出的基于分布式抗差卡尔曼滤波的多传感器融合算法进行了验证。试验结果表明,与单一传感器的测量值相比,所提出的算法能有效融合多个传感器的信息以提升目标的检测与跟踪精度,且鲁棒性较好。

Autonomous vehicles often use multiple sensors to detect and track surrounding targets.Howev-er,accurate multi-target detection and tracking remains a major challenge and difficulty in achieving autonomous driving due to the heterogeneous characteristics of sensors and complex driving environments.For the task of multi-target detection and tracking in Advanced Driver Assistance System(ADAS),a sensor configuration scheme based on a visual sensor and five millimeter-wave radars(1V5R)is used in this paper and a multi-sensor information fu-sion algorithm based on distributed robust Kalman filters is designed to realize accurate perception of surrounding targets.Firstly,considering different data characteristics of sensors,various Kalman filters such as linear Kalman filters and extended Kalman filters are adopted for data fusion and a 1V5R information fusion framework is built based on distributed Kalman filtering algorithm.Then,to reduce the impact of sensor dynamic error on fusion accu-racy,the robust estimation theory is introduced into the Kalman-weighted fusion,enabling real-time estimation and correction of dynamic error.Finally,the proposed multi-sensor information fusion algorithm is validated through sim-ulation and vehicle tests.The results show that compared to measurements from a single sensor,the proposed algo-rithm can robustly perform the task of information fusion of multiple sensors and improve the accuracy of detection and tracking,with good robustness.