Gated Recurrent Neural Network with Interaction Model of Multiple Targets for Traffic Radar Tracking

As urban transportation becomes more intelligent, traffic radar target tracking has emerged as a key research area. In complex traffic environments, clutter or obstacles cause variations in radar measurement data, leading to measurement interruptions or losses that challenge continuous and stable target tracking. Traditional Kalman filter (KF) methods and neural network (NN)-based approaches are limited in their ability to handle these variations. By integrating the strengths of KF and gated recurrent neural networks (GRNN), a novel GRNN is designed to leverage multi-target interaction for robust tracking in traffic scenarios. Specifically, the internal structure of the proposed GRNN is designed based on the computational processes of KF, and a memory factor is introduced to retain historical interaction information to surpass the limitations of traditional first-order Markov tracking methods while integrating the interaction information among multiple targets. The introduction of the KF framework brings excellent interpretability, a lightweight parameter scale, and low training data requirements. Experiments on the real-world dataset show that the proposed method significantly outperforms benchmark KF and advanced NN-based filtering methods in target tracking performance with measurement variations, ensuring continuous and stable tracking.