Roadside multi-sensor fusion based on adaptive extended Kalman filter

doi:10.3969/j.issn.1674-8484.2021.04.011

Abstract

Abstract:

Roadside perception is a component of cloud control cooperative vehicle infrastructure perception. This paper proposed a fusion method for roadside multi-sensors based on measurement noise Adaptive Extended Kalman Filter (AEKF) to improve the perception accuracy and stability of roadside sensors. The fusion of target-level data from heterogeneous sensors was realized based on the sensing results of roadside cameras, lidars, and millimeter-wave radars. An online acquisition method of measurement noise was used to detect the stability of the sensor measurement value, to generate correction coefficients for the measurement noise, and to adjust adaptively the measurement noise. This method was tested in a real vehicle. The results show that this multi-sensor fusion method improves the accuracy of lateral distance estimation by 9.7%, the longitudinal distance estimation accuracy by 5.4%, and the speed estimation accuracy by 26.6%, compared with a single sensor; With the estimation accuracy of the traditional Extended Kalman Filter (EKF) algorithm being improved by 44.9% in the horizontal distance, by 21.3% in the longitudinal distance, and by 64.4% in the speed; Therefore, the AEKF algorithm estimation accuracy of this paper is higher than that of the traditional EKF algorithm.

Key words: autonomous vehicles, roadside perception, multi-sensor fusion, adaptive filters, extended Kalman filter (EKF)

CLC Number:

TP273

WU Yimin, ZHENG Kaiyuan, GAO Bolin, CHEN Ming, WANG Yifeng. Roadside multi-sensor fusion based on adaptive extended Kalman filter[J]. Journal of Automotive Safety and Energy, 2021, 12(4): 522-527.

Figures/Tables 9

References 16

[1]	李克强, 戴一凡, 李升波, 等. 智能网联汽车(ICV)技术的发展现状及趋势[J]. 汽车安全与节能学报, 2017, 8(1):1-14.
	LI Keqiang, DAI Yifan, LI Shengbo, et al. State-of-the-art and technical trends of intelligent and connected vehicles[J]. J Automotive Safety Energy, 2017, 8(1):1-14. (in Chinese)
[2]	张长隆, 鲍海兴, 杜仙童, 等. 路侧感知在智能网联汽车中的应用与未来[J]. 人工智能, 2019(1):58-66.
	ZHANG Changlong, BAO Haixing, DU Xiantong, et al. The application and future of roadside perception in intelligent networked vehicles[J]. Artif Intell, 2019(1):58-66. (in Chinese)
[3]	Bleiholder J, Naumann F. Data fusion[J]. *ACM Computing Surveys*, 2009, 41(1):1-41.
[4]	WANG Zhangjing, WU Yu, NIU Qingqing. Multi-sensor fusion in automated driving: a survey[J]. *IEEE Access*, 2020, 8:2847-2868. doi: 10.1109/Access.6287639 URL
[5]	Alessandretti G, Broggi A, Cerri P. Vehicle and guard rail detection using radar and vison data fusion[J]. *Intell Transp Syst, IEEE Trans*, 2007, 8(1):95-105. doi: 10.1109/TITS.2006.888597 URL
[6]	Hoang M G, Denis B, Hrri J, et al. Bayesian fusion of GNSS, ITS-G5 and IR-UWB data for robust cooperative vehicular localization[J]. *Comptes Rendus Physique*, 2019, 20(3):218-227. doi: 10.1016/j.crhy.2019.03.004
[7]	王法胜, 鲁明羽, 赵清杰, 等. 粒子滤波算法[J]. 计算机学报, 2014, 37(8):1679-1694.
	WANG Fasheng, LU Mingyu, ZHAO Qingjie, et al. Particle Filtering Algorithm[J]. Chin J Computers, 2014, 37(8):1679-1694. (in Chinese)
[8]	Bergman M J, Delleur J W. Kalman filter estimation and perdiction of daily stream flows: I. review, algorithm, and simulation experiments[J]. *Jawra J Ame Water Resources Asso*, 2010, 21(5):815-825.
[9]	ZHAO Xiangmo, Kenan M, FEI Hui, et al. A cooperative vehicle-infrastructure based urban driving environment perception method using a D-S theory-based credibility map[J]. *Optik Int’l J Light Electron Optics, 2017, 138*:407-415.
[10]	黄文锦, 黄妙华. 激光雷达与路侧摄像头的双层融合协同定位[J]. 浙江大学学报(工学版), 2020, 54(7):1369-1379.
	HUANG Wenjin, HUANG Miaohua. Double-layer fusion of lidar and roadside camera for cooperative location[J]. J Zhejiang Univ (Engi Sci), 2020, 54(7):1369-1379. (in Chinese)
[11]	Al-Shabi M. Sigma-point filters in robotic applications[J]. *Intell Cont Automation*, 2015, 6(3):168-183.
[12]	黄彬. 基于多传感器融合的复杂行驶工况下多目标跟踪方法[D]. 北京: 清华大学, 2018.
	HUANG Bin. Multisensor-fusion-based multitarget tracking approach in complex driving situations[D]. Beijing: Tsinghua University, 2018. (in Chinese)
[13]	王向华, 覃征, 杨新宇, 等. 基于两次Kalman滤波的观测噪声自适应调整算法[J]. 系统工程与电子技术, 2010, 32(2):232-234.
	WANG Xianghua, QI Zheng, YANG Xinyu, et al. Adaptive algorithm for adjusting observation noises based on double-Kalman filter[J]. Syst Engi Electr, 2010, 32(2):232-234. (in Chinese)
[14]	杨登科. 不同插值方法对GPS时间序列的影响分析[J]. 全球定位系统, 2019, 44(5):66-69.
	YANG Dengke. Analysis of the influence of different interpolation methods on GPS time series[J]. GNSS World of China, 2019, 44(5):66-69. (in Chinese)
[15]	Jo K, Lee M, Kim J, et al. Tracking and behavior reasoning of moving vehicles based on roadway geometry constraints[J]. *IEEE Trans Intell Transp Syst*, 2017, 18(2):1-17. doi: 10.1109/TITS.6979 URL
[16]	Roecker J A, Mcgillem C D. Comparison of two-sensor tracking methods based on state vector fusion and measurement fusion[J]. *IEEE Trans Aerospace Elect Syst*, 1988, 24(4):447-449. doi: 10.1109/7.7186 URL

传感器	D_x / m	D_y / m	v / (m·s^-1)
相机	0.15	0.50	2.50
激光雷达	0.10	0.80	0.15
毫米波雷达	1.50	6.00	0.10

传感器	D_x / m	D_y / m	v / (m·s^-1)
相机	0.15	0.50	2.50
激光雷达	0.10	0.80	0.15
毫米波雷达	1.50	6.00	0.10

传感器或算法		D_x / m	D_y / m	v / （m·s^-1）
传感器	Camera	0.178 3	0.676 5	9.386 2
	Lidar	0.110 7	0.895 1	0.913 8
	Radar	1.471 1	6.115 6	0.450 1
算法	EKF	0.181 6	0.813 0	0.928 4
算法	AEKF	0.100 0	0.640 0	0.330 2

传感器或算法		D_x / m	D_y / m	v / （m·s^-1）
传感器	Camera	0.178 3	0.676 5	9.386 2
	Lidar	0.110 7	0.895 1	0.913 8
	Radar	1.471 1	6.115 6	0.450 1
算法	EKF	0.181 6	0.813 0	0.928 4
算法	AEKF	0.100 0	0.640 0	0.330 2

	横向距离	纵向距离	横向速度	纵向速度	横向加速度	纵向加速度	角速度
状态向量初值	23.880	-55.320	2.210	-6.718	-0.026	-0.301	0.017
状态协方差对角阵各元素初值	1	1	1	1	1	1	1
过程噪声 / 10^-7	0.01	0.001	10	0.01	100	0.1	1
相机量测噪声初值/10^-3	0.400	4.000	0.753	0.327	147	866	2.000
激光雷达量测噪声初值/10^-4	0.052	0.044	0.478	0.079	200	70	4.943
毫米波雷达量测噪声初值/ 10^-2	0.300	0.400	0.100	0.089	377.5	276.4	66.0
相机测量值阈值/ 10^-3	22	157	47	335	709	515	18
激光雷达测量值阈值/ 10^-3	0.261	1	4	3	470	177	7
毫米波雷达测量值阈值	0.020	0.018	0.020	0.005	1.334	0.288	0.032