基于ACT和Bi-LSTM的多模型交互跟踪算法

doi:10.12305/j.issn.1001-506X.2026.06.09

摘要/Abstract

摘要：

面对具有高机动特性无人机等气动目标，传统基于固定运动模型的目标跟踪算法在应对目标的复杂运动模式时，常因模型失配导致跟踪精度下降甚至目标丢失。针对这一问题，提出一种基于自适应转弯（adaptive coordinated turn，ACT）模型和双向长短时记忆（bidirectional long short-term memory，Bi-LSTM）网络的多模型交互跟踪算法。通过Bi-LSTM网络学习历史观测数据与目标运动状态之间的非线性关系，实现对目标转弯率的精确辨识；基于数据-模型混合驱动的思想，将神经网络的预测结果与交互多模型ACT（interacting multiple model-ACT，IMM-ACT）滤波算法相结合，有效提升算法在目标运动模式切换时的响应能力。实验结果表明，所提算法在目标运动模式多样化及高频切换场景下具有较高的跟踪精度，显著提升机动目标跟踪的稳定性和连续性，为机动目标跟踪提供了新的解决方案。

关键词: 机动目标跟踪, 双向长短时记忆网络, 自适应转弯模型, 交互多模算法

Abstract:

For aerodynamic targets with high maneuvering characteristics such as unmanned aerial vehicles, traditional target tracking algorithms based on fixed motion models often suffer from model mismatch when dealing with complex motion patterns. To address this issue, a multi-model interaction tracking algorithm is proposed based on adaptive coordinated turn （ACT） model and bidirectional long short-term memory （Bi-LSTM） network. The algorithm uses Bi-LSTM network to learn the nonlinear relationship between historical observation data and target motion state, and realizes the accurate identification of target turning rate. Based on the idea of hybrid data-model driven, the prediction results of neural network are combined with the interacting multiple model-ACT （IMM-ACT） filtering algorithm. It effectively improves the response ability of the algorithm when the target motion mode switches. The experimental results show that the proposed algorithm has higher tracking accuracy in the scenarios of target motion pattern diversification and high-frequency switching, which significantly improves the stability and continuity of maneuvering target tracking, and provides a solution to maneuvering target tracking.

Key words: maneuvering target tracking, bidirectional long short-term memory （Bi-LSTM） network, adaptive coordinated turn （ACT） model, interacting multiple model algorithm （IMM）

中图分类号:

TN 953

梁雨凡, 陈映, 李鑫, 窦凇耀. 基于ACT和Bi-LSTM的多模型交互跟踪算法[J]. 系统工程与电子技术, 2026, 48(6): 1869-1879.

Yufan LIANG, Ying CHEN, Xin LI, Songyao DOU. Multi-model interaction tracking algorithm based on ACT and Bi-LSTM[J]. Systems Engineering and Electronics, 2026, 48(6): 1869-1879.

图/表 18

表1

图1

图2

表2

图3

图4

表3

图5

表4

图6

图7

图8

表5

图9

图10

图11

图12

表6

参考文献 29

17	BECKER S, HUG R, HUBNER W, et al. An RNN-based IMM filter surrogate[C]//Proc. of the Scandinavian Conference on Image Analysis, 2019: 387−398.
18	DENG L, LI D, LI R. Improved IMM algorithm based on RNNs[C]//Journal of Physics: Conference Series. IOP Publishing, 2020, 1518(1): 012055.
19	马云红, 李欣怡, 王泽霖, 等. 基于数据驱动的机动目标跟踪[J]. 航空科学技术, 2024, 35 (4): 97- 103.
	MA Y H, LI X Y, WANG Z L, et al. Maneuvering target tracking based on data-driven[J]. Aeronautical Science and Technology, 2024, 35 (4): 97- 103.
20	LIU B, SHI Q, HAN P. Short-term 4D trajectory prediction method based on LSTM-IMM[C]// Proc. of the IEEE Conference on Digital Avionics Systems, 2022.
21	TIAN W, FANG L, LI W, et al. Deep-learning-based multiple model tracking method for targets with complex maneuvering motion[J]. Remote Sensing, 2022, 14 (14): 3276. doi: 10.3390/rs14143276
22	JOUABER S, BONNABEL S, VELASCO-FORERO S, et al. NNAKF: a neural network adapted Kalman filter for target tracking[C]//Proc. of the ICASSP IEEE International Conference on Acoustics, Speech and Signal Processing, 2021: 4075−4079.
23	SUN M, MA Z, LI Y. Maneuvering target tracking using IMM Kalman filter aided by Elman neural network[C]//Proc. of the 7th International Conference on Intelligent Human-Machine Systems and Cybernetics, 2015, 1: 144−148.
24	王纯鑫. 结合数据驱动的雷达目标跟踪方法研究[D]. 西安: 西安电子科技大学, 2022.
	WANG C X. Research on radar target tracking method combined with data-driven [D]. Xi’an: Xidian University, 2022.
25	SHRESTHA A, LI H, LE KERNEC J, et al. Continuous human activity classification from FMCW radar with Bi-LSTM networks[J]. IEEE Sensors Journal, 2020, 20 (22): 13607- 13619. doi: 10.1109/JSEN.2020.3006386
26	BLOM H A P, BAR-SHALOM Y. The interacting multiple model algorithm for systems with Markovian switching coefficients[J]. IEEE Trans. on Automatic Control, 1988, 33 (8): 780- 783. doi: 10.1109/9.1299
1	WU Y, LIM J, YANG M H. Online object tracking: a benchmark[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2013: 2411−2418.
2	丁鹭飞, 耿富录, 陈建春. 雷达原理[M]. 陕西: 西安电子科技大学出版社, 2013.
	DING L F, GENG F L, CHEN J C. Radar principles[M]. Xi’an: Xidian University Press, 2013.
3	周宏仁, 敬忠良, 王德培. 机动目标跟踪[M]. 北京: 国防工业出版社, 1991.
	ZHOU H R, JING Z L, WANG D P. Maneuvering target tracking[M]. Beijing: National Defense Industry Press, 1991.
4	姜尚, 魏波, 梁伟阁, 等. 考虑齿隙的多约束导引控制一体化设计方法[J]. 系统工程与电子技术, 2022, 44 (4): 1318- 1328.
	JIANG S, WEI B, LIANG W G, et al. Integrated guidance and control design method with multiple constraints and backlash[J]. Systems Engineering and Electronics, 2022, 44 (4): 1318- 1328.
5	梁津鑫, 唐奇, 崔颢, 等. 强跟踪容积卡尔曼滤波在空空导弹制导中的应用[J]. 航空兵器, 2024, 31 (5): 82- 87. doi: 10.12132/ISSN.1673-5048.2024.0004
	LIANG J X, TANG Q, CUI H, et al. Application of strong tracking cubature Kalman filter in air-to-air missile guidance[J]. Aero Weaponry, 2024, 31 (5): 82- 87. doi: 10.12132/ISSN.1673-5048.2024.0004
6	LI S J, LEI H M, SHAO L, et al. Multiple model tracking for hypersonic gliding vehicles with aerodynamic modeling and analysis[J]. IEEE Access, 2019, 7 (1): 28011- 28018.
7	冯树林, 郭杰, 唐胜景. 一种改进的高超声速滑翔目标跟踪方法[J]. 飞行力学, 2023, 41 (5): 74- 80.
	FENG S L, GUO J, TANG S J. An improved hypersonic gliding target tracking method[J]. Flight Dynamics, 2023, 41 (5): 74- 80.
8	PETRIDIS V, KEHAGIAS A. A multi-model algorithm for parameter estimation of time varying nonlinear systems[J]. Automatica, 1998, 34 (4): 469- 475. doi: 10.1016/S0005-1098(97)00203-3
9	陈维义, 何凡, 刘国强, 等. 一种自适应变结构交互式多模型算法[J]. 火力与指挥控制, 2023, 48 (12): 177- 183. doi: 10.3969/j.issn.1002-0640.2023.12.027
	CHEN W Y, HE F, LIU G Q, et al. An adaptive variable structure interacting multiple model algorithm[J]. Fire Control and Command Control, 2023, 48 (12): 177- 183. doi: 10.3969/j.issn.1002-0640.2023.12.027
10	MONDAL B, PANAKKAL V P, VELMURUGAN R, et al. IMM filtering with randomized turn rate to track maneuvering target[C]// Proc. of the IEEE Aerospace Conference, 2024.
11	NDIKUMANA A, TRAN N H, KIM D H, et al. Deep learning based caching for self-driving cars in multi-access edge computing[J]. IEEE Trans. on Intelligent Transportation Systems, 2020, 22 (5): 2682- 2877.
12	王俊, 郑彤, 雷鹏, 等. 深度学习在雷达中的研究综述[J]. 雷达学报, 2018, 7 (4): 395- 411.
	WANG J, ZHENG T, LEI P, et al. Survey on deep learning in radar[J]. Journal of Radars, 2018, 7 (4): 395- 411.
13	LIU Z, WANG Z, YANG Y, et al. A data-driven maneuvering target tracking method aided with partial models[J]. IEEE Trans. on Vehicular Technology, 2023, 73 (1): 414- 425.
14	朱洪峰, 熊伟, 崔亚奇, 等. 新的自适应转弯模型的IMM算法研究[J]. 计算机工程与应用, 2019, 55 (17): 252- 258. doi: 10.3778/j.issn.1002-8331.1805-0351
	ZHU H F, XIONG W, CUI Y Q, et al. Research on IMM algorithm of new adaptive turning model[J]. Computer Engineering and Applications, 2019, 55 (17): 252- 258. doi: 10.3778/j.issn.1002-8331.1805-0351
15	GAO C, YAN J K, ZHOU S H, et al. Long short-term memory-based deep recurrent neural networks for target tracking[J]. Information Sciences, 2019, 502, 279- 296. doi: 10.1016/j.ins.2019.06.039
16	LIU J, WANG Z, XU M. Deep MTT: a deep learning maneuvering target-tracking algorithm based on bidirectional LSTM network[J]. Information Fusion, 2020, 53, 289- 304. doi: 10.1016/j.inffus.2019.06.012
27	张彪, 孙超军, 徐嘉辉, 等. 自适应CT模型的强跟踪CKF算法[J]. 舰船电子工程, 2024, 44 (2): 104- 108. doi: 10.3969/j.issn.1672-9730.2024.02.022
	ZHANG B, SUN C J, XU J H, et al. Strong tracking CKF algorithm based on adaptive CT model[J]. Ship Electronic Engineering, 2024, 44 (2): 104- 108. doi: 10.3969/j.issn.1672-9730.2024.02.022
28	杨建秀. 无人机视角下车辆目标检测的算法研究[D]. 西安: 西安电子科技大学, 2021.
	YANG J X. Research on algorithm of vehicle target detection from human-machine perspective [D]. Xi’an: Xidian University, 2021.
29	秦雷, 李君龙, 周荻. 基于AGIMM的临近空间机动目标跟踪滤波算法[J]. 系统工程与电子技术, 2015, 37 (5): 1009- 1014.
	QIN L, LI J L, ZHOU D. Filtering algorithm for maneuvering target tracking in near space based on AGIMM[J]. Systems Engineering and Electronics, 2015, 37 (5): 1009- 1014.

参数	参数值
CV目标数	1900
CT目标数	9700
CA目标数	6400
跟踪时长/s	300
采样间隔T/s	1
距离范围/km	$[0,75]$
速度范围/（m/s）	$\left[ {0,340} \right]$
转弯率范围/（°/s）	$[ - 10,10]$
距离噪声标准差/m	$[2,5]$
加速度噪声标准差/（${\text{m/}}{{\text{s}}^2}$）	$[2,8]$
方位角噪声标准差/（°）	$[0.1,1]$

参数	参数值	说明
input_size	2	每个时间步的输入维度
hidden_size	128	LSTM隐藏状态的维度
output_size	n	分类的类别数
num_layers	2	LSTM的层数
dropout	0.3	随机丢弃比例
batch_size	64	每次训练的样本数量
lr	0.001	Adam优化器的学习率

轨迹编号	初始状态	起始时刻/s	终止时刻/s
A	${\left[ {1 \times {{10}^4},400,3 \times {{10}^4}, - 100} \right]^{\text{T}}}$	1	100
B	${\left[ {1.2 \times {{10}^4}, - 410.1,5.3 \times {{10}^4}, - 42.8} \right]^{\text{T}}}$	101	150
C	${\left[ {1.2 \times {{10}^4},409.2,6.5 \times {{10}^4}, - 50.5} \right]^{\text{T}}}$	151	190

参数	参数值
跟踪时长$T$/s	190
采样间隔${T_{\text{s}}}$/s	1
覆盖概率${P_{{\mathrm{target}}}}$/%	70
最大转弯率${\omega _{\max }}/(^\circ )$	10

算法	位置/m	速度/（m/s）
ACT	34.87	14.85
AGIMM	15.15	8.16
BiLSTM-IMM-ACT	11.92	3.41