Abstract
In this study, we implemented learning-based and hybrid methodologies for maritime obstacle tracking leveraging RADAR (RAdio Detection And Ranging) data. The learning-based method employs a deep learning algorithm to predict the state of the obstacle. It is compared with the EKF-based method, which utilizes an extended Kalman filter extensively used in obstacle tracking. Moreover, we propose a hybrid method that combines the strengths of both the learning-based and EKF-based methods to compensate for their respective shortcomings. Experimental RADAR data from field tests were utilized for validation, illustrating that the hybrid method improves tracking accuracy compared to the EKF-based approach.