基于双支协同滤波网络的目标跟踪方法

张文安; 乔小龙; 林安迪; 杨旭升

doi:10.16383/j.aas.c250590

基于双支协同滤波网络的目标跟踪方法

doi: 10.16383/j.aas.c250590 cstr: 32138.14.j.aas.c250590

张文安^{1, 2,},
乔小龙^{1, 2,},
林安迪^{1, 2,},
杨旭升^{1, 2,}

1.
浙江工业大学信息工程学院杭州 310023
2.
浙江省复杂系统智能感知与控制重点实验室杭州 310023

基金项目: 国家自然科学基金(U25A20456, 62473335, W2421117), 杭州市科技发展计划项目(2022AIZD0080)资助

详细信息

作者简介:
张文安：浙江工业大学信息工程学院教授. 主要研究方向为多源信息融合估计和网络化系统. E-mail: wazhang@zjut.edu.cn

乔小龙：浙江工业大学信息工程学院硕士研究生. 主要研究方向为多源信息融合估计和深度学习. E-mail: 211123030055@zjut.edu.cn

林安迪：浙江工业大学信息工程学院博士研究生. 主要研究方向为多源信息融合估计. E-mail: 201706061126@zjut.edu.cn

杨旭升：浙江工业大学信息工程学院副教授. 主要研究方向为多源信息融合估计和目标定位. 本文通信作者. E-mail: xsyang@zjut.edu.cn

计量
- 文章访问数: 237
- HTML全文浏览量: 101
- PDF下载量: 30
- 被引次数: 0
出版历程
- 收稿日期: 2025-10-31
- 录用日期: 2025-12-31
- 网络出版日期: 2026-03-17
- 刊出日期: 2026-04-20

A Target Tracking Method Based on Dual-Branch Collaborative Filtering Network

ZHANG Wen-An^{1, 2
,},
QIAO Xiao-Long^{1, 2
,},
LIN An-Di^{1, 2
,},
YANG Xu-Sheng^{1, 2
,}

1.
College of Information Engineering, Zhejiang University of Technology, Hangzhou 310023
2.
Zhejiang Key Laboratory of Intelligent Perception and Control for Complex Systems, Hang-zhou 310023

Funds: Supported by National Natural Science Foundation of China (U25A20456, 62473335, W2421117) and Hangzhou Science and Technology Development Plan Project (2022AIZD0080)

More Information

Author Bio:
ZHANG Wen-An　Professor at the College of Information Engineering, Zhejiang University of Technology. His research interests include multi-source information fusion estimation and networked systems

QIAO Xiao-Long　Master student at the College of Information Engineering, Zhejiang University of Technology. His research interests include multi-source information fusion estimation and deep learning

LIN An-Di　Ph.D. candidate at the College of Information Engineering, Zhejiang University of Technology. His main research interest is multi-source in-formation fusion estimation

YANG Xu-Sheng　Associate professor at the College of Information Engineering, Zhejiang University of Technology. His research interests include multi-source information fusion estimation and target positioning. Corresponding author of this paper

摘要

摘要: 针对时序−状态相关性提取不足引起的目标跟踪性能下降问题, 提出一种基于双支协同滤波网络(DBCF-Net)的目标跟踪方法. 首先, 为实现运动模型和过程噪声参数的动态调整, 分别设计非马尔科夫信息网络和状态相关信息网络, 以学习运动目标状态演化过程中的时序依赖性及其状态变量间的局部相关性; 其次, 设计一种基于最大均值差异的网络权重协同更新机制, 通过差异化分支网络输出特征来增强分支网络间的学习互补性, 从而提升DBCF-Net对未知运动模式的适应能力; 进而, 融合贝叶斯滤波与神经网络的优势, 将无偏量测转换引入DBCF-Net, 以增强目标跟踪的鲁棒性; 最后, 通过目标跟踪实验验证了DBCF-Net的有效性.
- 状态估计 /
- 卡尔曼滤波 /
- 目标跟踪 /
- 滤波网络
Abstract: To address the performance degradation in target tracking caused by insufficient extraction of temporal-state correlations, a dual-branch collaborative filtering network (DBCF-Net) based target tracking method is proposed. First, to achieve dynamic adjustment of motion model and process noise parameters, a non-Markov information network and a state-related information network are designed separately to learn the temporal dependencies in the motion target state evolution process and the local correlations among state variables. Second, a network weight collaborative update mechanism based on maximum mean discrepancy is designed to enhance learning complementarity between the branch networks by differentiating their output features, thereby improving the adaptability of DBCF-Net to unknown motion patterns. Furthermore, leveraging the strengths of Bayesian filtering and neural networks, unbiased measurement transformation is introduced into DBCF-Net to enhance the robustness of target tracking. Finally, target tracking experiments validate the effectiveness of DBCF-Net.
- state estimation /
- Kalman filter /
- target tracking /
- filtering network

HTML全文

图 1 目标运动的时序相关性和局部相关性

Fig. 1 Temporal correlation and local correlation of target motion

下载: 全尺寸图片幻灯片

图 2 双支协同滤波网络框图

Fig. 2 Dual-branch collaborative network block diagram

下载: 全尺寸图片幻灯片

图 3 双支协同网络内部框图

Fig. 3 Dual-branch collaborative network internal block diagram

下载: 全尺寸图片幻灯片

图 4 训练数据轨迹图

Fig. 4 Training data trajectory chart

下载: 全尺寸图片幻灯片

图 5 6条轨迹的跟踪结果, 其中放大的子图中包含30个采样点的轨迹片段(主图中每隔2.5 s (25个采样点)标记一次采样点, 子图中每隔0.5 s标记一次采样点)

Fig. 5 The tracking results of six trajectories, where the enlarged subplot contains 30 sampled trajectory segments (In the main plot, sampling points are marked at intervals of 2.5 s (corresponding to 25 sampling points), while in the subplot, they are marked every 0.5 s)

下载: 全尺寸图片幻灯片

图 6 6条测试轨迹的目标状态估计的均方根误差

Fig. 6 RMSE of the target state estimation for the six test trajectories

下载: 全尺寸图片幻灯片

图 7 消融实验跟踪结果可视化图

Fig. 7 Visualization of the tracking results obtained in the ablation experiments

下载: 全尺寸图片幻灯片

表 1 测试轨迹运动参数

Table 1 Test trajectory maneuver parameters

轨迹序号	初始状态	第1段	第2段	第3段
1	$ [-17\;000.0\;{\mathrm{m}},\;2\;600.0\;{\mathrm{m}},\;200.0\;{\mathrm{m}}/{\mathrm{s}},\;120.0\;{\mathrm{m}}/{\mathrm{s}}] $	$ 20\;\text{s},\;\text{CV} $	$ 25\;\text{s},\;\text{CT},\;\omega=3.6\;({\text{°}}) /\text{s} $	$ 30\;\text{s},\;\text{CT},\;\omega=-6.4\;({\text{°}}) /\text{s} $
2	$ [-6\;860.0\;{\mathrm{m}},\;24\;320.0\;{\mathrm{m}},\;90.0\;{\mathrm{m}}/{\mathrm{s}},\;-130.0\;{\mathrm{m}}/{\mathrm{s}}] $	$ 25\;\text{s},\;\text{CT},\;\omega=1.0\;({\text{°}}) /\text{s} $	$ 25\;\text{s},\;\text{CT},\;\omega=-1.6\;({\text{°}}) /\text{s} $	$ 25\;\text{s},\;\text{CT},\;\omega=-6.4\;({\text{°}}) /\text{s} $
3	$ [17\;155.0\;{\mathrm{m}},\;-9\;300.0\;{\mathrm{m}},\;-169.0\;{\mathrm{m}}/{\mathrm{s}},\;140.0\;{\mathrm{m}}/{\mathrm{s}}] $	$ 10\;\text{s},\;\text{CV} $	$ 50\;\text{s},\;\text{CT},\;\omega=8.00\;({\text{°}}) /\text{s} $	$ 15\;\text{s},\;\text{CV} $
4	$ [13\;345.0\;{\mathrm{m}},\;-11\;300.0\;{\mathrm{m}},\;69.0\;{\mathrm{m}}/{\mathrm{s}},\;140.0\;{\mathrm{m}}/{\mathrm{s}}] $	$ 25\;\text{s},\;\text{CV} $	$ 30\;\text{s},\;\text{CT},\;\omega=-7.0\;({\text{°}}) /\text{s} $	$ 20\;\text{s},\;\text{CT},\;\omega=6.48\;({\text{°}}) /\text{s} $
5	$ [19\;134.0\;{\mathrm{m}},\;19\;144.0\;{\mathrm{m}},\;-235.0\;{\mathrm{m}}/{\mathrm{s}},\;-33.0\;{\mathrm{m}}/{\mathrm{s}}] $	$ 20\;\text{s},\;\text{CT},\;\omega=6.08\;({\text{°}}) /\text{s} $	$ 30\;\text{s},\;\text{CV} $	$ 25\;\text{s},\;\text{CT},\;\omega=-9.01\;({\text{°}}) /\text{s} $
6	$ [9\;360.0\;{\mathrm{m}},\;-8\;740.0\;{\mathrm{m}},\;-140.0\;{\mathrm{m}}/{\mathrm{s}},\;-1.0\;{\mathrm{m}}/{\mathrm{s}}] $	$ 20\;\text{s},\;\text{CT},\;\omega=9.08\;({\text{°}}) /\text{s} $	$ 30\;\text{s},\;\text{CT},\;\omega=-8.1\;({\text{°}}) /\text{s} $	$ 25\;\text{s},\;\text{CT},\;\omega=1.08\;({\text{°}}) /\text{s} $
注: CT: Constant turn.

下载: 导出CSV

表 2 不同方法在测试轨迹上的平均均方根误差(ARMSE)

Table 2 The ARMSE of states for different methods on the test trajectory

方法	参数	轨迹1	轨迹2	轨迹3	轨迹4	轨迹5	轨迹6
IMM-EKF	位置(m)	4.872	5.208	12.942	4.942	5.969	4.236
IMM-EKF	速度(m/s)	9.606	6.569	21.949	8.336	11.263	8.919
IMM-UKF	位置(m)	5.089	5.267	5.564	5.082	6.310	4.437
IMM-UKF	速度(m/s)	10.149	6.689	11.320	8.707	12.177	9.404
DeepMTT	位置(m)	6.061	5.576	7.240	4.889	9.473	5.797
DeepMTT	速度(m/s)	3.676	4.493	6.904	4.400	7.595	6.045
KalmanNet	位置(m)	11.302	14.067	6.641	5.863	17.151	4.977
KalmanNet	速度(m/s)	12.279	13.168	15.105	13.652	14.708	9.856
DBCF-Net	位置(m)	2.678	4.400	3.339	3.365	4.364	2.682
DBCF-Net	速度(m/s)	3.806	4.430	4.900	3.956	5.103	3.938
注: 加粗字体表示最优结果.

下载: 导出CSV

表 3 消融实验测试轨迹运动参数

Table 3 Test trajectory maneuver parameters of ablation experiment

轨迹序号	初始状态	第1段	第2段	第3段
1	[−19280.0 m, 18250.0 m, 180.0 m/s, 50.0 m/s]	$ 5\;\text{s},\;\text{CV} $	$ 20\;\text{s},\;\text{CT},\; $ $ \omega=-9.0\;({\text{°}}) /\text{s} $	$ 15\;\text{s},\;\text{CT},\; $ $ \omega= 8.4\;({\text{°}}) /\text{s} $
2	[−16900.0 m, 15500.0 m, 220.0 m/s, 300.0 m/s]	$ 5\;\text{s},\;\text{CV} $	$ 15\;\text{s},\;\text{CT},\; $ $ \omega=5.0\;({\text{°}}) /\text{s} $	$ 20\;\text{s},\;\text{CT},\; $ $ \omega=-3.4\;({\text{°}}) /\text{s} $

下载: 导出CSV

表 4 消融实验测试轨迹ARMSE值

Table 4 ARMSE of ablation experiment test trajectory

方法	参数	轨迹1	轨迹2
DBCF-Net	位置(m)	5.106	5.317
DBCF-Net	速度(m/s)	6.161	7.265
Single 1	位置(m)	6.758	8.169
Single 1	速度(m/s)	8.908	9.801
Single 2	位置(m)	6.920	8.409
Single 2	速度(m/s)	10.813	7.631
No MMD	位置(m)	7.233	9.209
No MMD	速度(m/s)	9.666	9.189

下载: 导出CSV

参考文献(24)

[1]	Cortina E, Otero D, Attellis C E. Maneuvering target tracking using extended Kalman filter. IEEE Transactions on Aerospace and Electronic Systems, 1991, 27(1): 155−158 doi: 10.1109/7.68158
[2]	Julier S J, Uhlmann J K. Unscented filtering and nonlinear estimation. Proceedings of the IEEE, 2004, 92(3): 401−422
[3]	Solaiman S, Alsuwat E, Alharthi R. Simultaneous tracking and recognizing drone targets with millimeter-wave radar and convolutional neural network. Applied System Innovation, 2023, 6(4): Article No. 68 doi: 10.3390/asi6040068
[4]	Yan B, Wei Y, Liu S, Huang W, Feng R, Chen X. A review of current studies on the unmanned aerial vehicle-based moving target tracking methods. Defence Technology, 2025, 51: 201−219 doi: 10.1016/j.dt.2025.01.013
[5]	Alhafnawi M, Bany S H, Masadeh A, Al-Obiedollah H, Ayyash M, El-Khazali R, et al. A survey of indoor and outdoor UAV-based target tracking systems: Current status, challenges, technologies, and future directions. IEEE Access, 2023, 11: 68324−68339 doi: 10.1109/ACCESS.2023.3292302
[6]	Yasmeen A, Daescu O. Recent research progress on ground-to-air vision-based anti-UAV detection and tracking methodologies: A review. Drones, 2025, 9(1): Article No. 58 doi: 10.3390/drones9010058
[7]	Yang Y, Moran B, Wang X, Brown T C, Williams S, Pan Q. Experimental analysis of a game-theoretic formulation of target tracking. Automatica, 2020, 114: 1−10
[8]	Magill D. Optimal adaptive estimation of sampled stochastic processes. IEEE Transactions on Automatic Control, 1965, 10(4): 434−439 doi: 10.1109/TVT.2020.2976095
[9]	Blom H A P, Bar-Shalom Y. The interacting multiple model algorithm for systems with Markovian switching coefficients. IEEE Transactions on Automatic Control, 1988, 33(8): 780−783 doi: 10.1109/9.1299
[10]	Daeipour E, Bar-Shalom Y. IMM tracking of maneuvering targets in the presence of glint. IEEE Transactions on Aerospace and Electronic Systems, 1998, 34(3): 996−1003 doi: 10.1109/7.705913
[11]	Wu W, Cheng P. A nonlinear IMM algorithm for maneuvering target tracking. IEEE Transactions on Aerospace and Electronic Systems, 1994, 30(3): 875−886 doi: 10.1109/7.303756
[12]	Li W, Jia Y, Du J, Yu F. Gaussian mixture PHD smoother for jump markov models in multiple maneuvering targets tracking. In: Proceedings of the American Control Conference. San Francisco, USA: IEEE, 2011. 3024−3029
[13]	Xu L, Li X R. Hybrid grid multiple-model estimation with application to maneuvering target tracking. In: Proceedings of the 13th International Conference on Information Fusion. Edinburgh, UK: IEEE, 2010. 1−7
[14]	Liu J, Wang Z, Xu M. DeepMTT: A deep learning maneuvering target-tracking algorithm based on bidirectional LSTM network. Information Fusion, 2020, 53: 289−304 doi: 10.1016/j.inffus.2019.06.012
[15]	Yang X, Qiao D. Attention-based bidirectional LSTM network for target tracking. In: Proceedings of IEEE International Conference on Electronic Technology, Communication and Information. Changchun, China: IEEE, 2021. 151−156
[16]	Zhai B, Yi W, Li M, Ju H, Kong L. Data-driven XGBoost-based filter for target tracking. In: Proceedings of the International Radar Conference. Nanjing, China: IET, 2019. 6683−6687
[17]	张文安, 林安迪, 杨旭升, 俞立, 杨小牛. 融合深度学习的贝叶斯滤波综述. 自动化学报, 2024, 50(8): 1502−1516 doi: 10.16383/j.aas.c230457 Zhang Wen-An, Lin An-Di, Yang Xu-Sheng, Yu Li, Yang Xiao-Niu. A survey on Bayesian filtering with deep learning. Acta Automatica Sinica, 2024, 50(8): 1502−1516 doi: 10.16383/j.aas.c230457
[18]	Chen C, Lu C X, Wang B, Trigoni N, Markham A. DynaNet: Neural Kalman dynamical model for motion estimation and prediction. IEEE Transactions on Neural Networks and Learning Systems, 2021, 32(12): 5479−5491 doi: 10.1109/TNNLS.2021.3112460
[19]	杨旭升, 李福祥, 胡佛, 张文安. 基于肌电−惯性融合的人体运动估计: 高斯滤波网络方法. 自动化学报, 2024, 50(5): 991−1000 doi: 10.16383/j.aas.c230581 Yang Xu-Sheng, Li Fu-Xiang, Hu Fo, Zhang Wen-An. Human motion estimation based on EMG-inertial fusion: A Gaussian filtering network approach. Acta Automatica Sinica, 2024, 50(5): 991−1000 doi: 10.16383/j.aas.c230581
[20]	Revach G, Shlezinger N, Ni X, Escoriza A L, van Sloun R J G, Eldar Y C. KalmanNet: Neural network aided Kalman filtering for partially known dynamics. IEEE Transactions on Signal Processing, 2022, 70: 1532−1547 doi: 10.1109/TSP.2022.3158588
[21]	Choi G, Park J, Shlezinger N, Eldar Y C, Lee N. Split KalmanNet: A robust model-based deep learning approach for state estimation. IEEE Transactions on Vehicular Technology, 2023, 72(9): 12326−12331 doi: 10.1109/TVT.2023.3270353
[22]	Escoriza A L, Revach G, Shlezinger N, van Sloun R J G. Data-driven Kalman-based velocity estimation for autonomous racing. In: Proceedings of the IEEE International Conference on Autonomous Systems. Montreal, QC, Canada: IEEE, 2021. 1−5
[23]	杨旭升, 王雪儿, 汪鹏君, 张文安. 基于渐进无迹卡尔曼滤波网络的人体肢体运动估计. 自动化学报, 2023, 49(8): 1723−1731 doi: 10.16383/j.aas.c220523 Yang Xu-Sheng, Wang Xue-Er, Wang Peng-Jun, Zhang Wen-An. Estimation of human limb motion based on progressive unscented Kalman filter network. Acta Automatica Sinica, 2023, 49(8): 1723−1731 doi: 10.16383/j.aas.c220523
[24]	Ding X, Guo Y, Ding G, Han J. ACNet: Strengthening the kernel skeletons for powerful CNN via asymmetric convolution blocks. In: Proceedings of IEEE/CVF International Conference on Computer Vision (ICCV). Seoul, South Korea: 2019. 1911−1920