针对小车状态感知的卡尔曼滤波多传感器融合算法  

Research on fusion algorithm of Kalman filter multi-sensors for car state perception

作　　者：任其亮[1] 程昊东 REN Qiliang;CHENG Haodong(College of Traffic and Transportation,Chongqing Jiaotong University,Chongqing 400074,China)

机构地区：[1]重庆交通大学交通运输学院,重庆400074

出　　处：《重庆理工大学学报（自然科学）》2022年第11期176-182,共7页Journal of Chongqing University of Technology：Natural Science

基　　金：国家社会科学基金项目(21BJY038)。

摘　　要：为解决视觉传感器和雷达获取小车周边环境信息时的数据漂移问题,引入改进扩展卡尔曼滤波对多传感器信息进行融合处理,并将其应用在环境感知模型中。根据小车运动特性建立恒定转率和速度模型,以小车运动实时状态量为输入量构建状态转移函数。以视觉传感器和雷达自带的里程计坐标系建立传感器测量模型,与小车运动模型相结合追踪小车实时运动状态,对小车运动数据进行改进扩展卡尔曼滤波交叉融合,根据雷达和视觉传感器测量精确度的差距,时刻更新相应的噪声协方差矩阵。用搭载ZED双目摄像头及雷达的四轮非全向小车模拟实验,并与传统扩展卡尔曼滤波实验结果比较验证模型的有效性。结果表明:使用改进扩展卡尔曼滤波处理后小车状态曲线更加平滑,数据更为精确,几乎不存在零点漂移。相较改进扩展卡尔曼滤波,传统扩展卡尔曼滤波的最大误差增大了55%,平均误差增大了44%,均方误差增大了17%,且运行处理数据时间更长。In order to solve the problem of data drift when the vision sensor and the radar obtain the surrounding environment information of the car,this paper introduces an improved extended Kalman filter to fuse the multi-sensor information,and applies it to the environment perception model.Firstly,the constant rotation rate and the speed model are established according to the motion characteristics of the car,and the state transfer function is constructed with the real-time state of the car motion as the input.Secondly,the sensor measurement model is constructed in the odometer coordinate system of the visual sensor and the radar.Combined with the car motion model to track the real-time motion state,the model improves the car motion data,expands the Kalman filter fusion,and timely updates the corresponding noise covariance matrix according to the measurement accuracy gap between the radar and the visual sensor.Finally,the simulation experiment of a four-wheel non-omnidirectional car equipped with ZED binocular cameras and radar is carried out,and the effectiveness of the model is verified through the comparison with the experimental results of traditional extended Kalman.The results show that,after using the improved extended Kalman filter,the car state curve is smoother,the data is more accurate,and there is almost no zero drift.Compared with the improved extended Kalman filter,the maximum error of the traditional algorithm increases by 55%,the average error increases by 44%,the mean square error increases by 17%,and the data processing time is longer.

关 键 词：非全向小车 扩展卡尔曼滤波 环境感知 视觉传感器 雷达 

分 类 号：TB24[一般工业技术—工程设计测绘]