A BiGRU Based Adaptive Gain Estimation for Radar Multi-target Tracking

The currently available multi-target tracking algorithms were developed based on ideal tracking settings, which are unsuitable for actual combat situations. To handle the challenge of unknown measurement noise and low detection probability, this paper presents a novel adaptive gain estimation (AGE) approach with a survival likelihood estimation method. The former utilizes the bidirectional gate recurrent unit (BiGRU) to assign weights to both measurement and prediction in the absence of priori information. The latter leverages a binary classification network to determine the likelihood of target survival. Furthermore, the AGE can be seamlessly integrated with prediction and data association modules to form an end-to-end model known as MTT-AGE (Multi-target Tracking with Adaptive Gain Estimation). The results of MIT trajectory dataset, simulated scenarios and real-world data confirm the efficiency and stability of the MTT-AGE. Furthermore, the ablation experiments are conducted to verify the effectiveness of AGE based on MIT trajectory dataset and simulated scenarios where there is only a single objective.