Hybrid-Driven Multiple Target Tracking Using Pulse Description Word From Targets' Radar Sensors

• Published in: IEEE transactions on instrumentation and measurement Vol. 74; pp. 1 - 17
• Main Authors: Li, Jimin, Wu, Panlong, Li, Xingxiu, Liu, Bingzhuo, He, Shan, Zhao, Ruohan
• Format: Journal Article
• Language: English
• Published: New York IEEE 2025; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Artificial neural networks; Classification algorithms; Computational modeling; Deep learning; Dynamic models; Feature extraction; Filtering algorithms; Hybrid driven; joint probabilistic data association; Kalman filters; Kinematics; Machine learning; Maneuvering targets; multiple maneuvering target tracking (MMTT); Multiple target tracking; Position measurement; Probability; Probability density functions; pulse descriptor words; radar countermeasure; Radar tracking; Simulation; Statistical analysis; Target tracking; Tracking; Trajectory
• ISSN: 0018-9456; 1557-9662
• DOI: 10.1109/TIM.2025.3569903

In the domain of radar countermeasures, existing multitarget tracking algorithms often struggle with tracking multiple maneuvering targets in dense scenarios. To address this issue, this article proposes a hybrid-driven algorithm that integrates both data and models. Initially, an enhanced joint probabilistic data association (JPDA) framework is proposed by incorporating target position and attribute measurements. In this algorithm, the position and attribute measurements associated with the targets are modeled as Gaussian and multi-Bernoulli distributions, respectively. The probability of each association event is then derived using the probability density function of the joint distribution. Subsequently, a deep learning feature extraction network [convolutional neural network-bidirectional long short-term memory (CNN-BiLSTM)] is designed to extract target attribute measurement features from multiple dimensions. This network generates the probability density function of the multi-Bernoulli distribution, thereby facilitating a data-driven association task. Finally, the associated target measurements are input into an interacting multiple model (IMM) unbiased converted Kalman filter to estimate the states of maneuvering targets. The proposed algorithm's effectiveness, convergence, robustness, and practical applicability are validated through simulation scenarios involving multiple maneuvering target tracking (MMTT), utilizing both the F16 six-degree-of-freedom dynamic model and a kinematic model.