Particle filter based target tracking from X-band nautical radar images

Abstract

In this thesis, two particle filter (PF) based visual tracking approaches are designed for maneuvering target tracking from X-band nautical radar images: a PF-only based approach and a combined particle-Kalman filters (PF-KF) based approach. Unlike existing Kalman filter (KF) based target tracking algorithms used by nautical radar, these two proposed tracking methods both employ a kernel-based histogram model to represent the target in the radar image, and a Bhattacharyya coefficient based similar- ity distance between reference and candidate target models to provide the likelihood function for the particle filtering. However, the PF-KF method applies a sampling importance resampling (SIR) particle filter to obtain preliminary target positions, and then a Kalman filter to derive refined target positions and velocities. Moreover, several strategies are also proposed to improve the tracking accuracy and stability. These strategies include an enhanced reference target model construction method, updating reference target model, and adaptive KF for maneuver. Comparison of the target information obtained by the proposed PF-KF method from various field X-band nautical radar image sequences with those measured by GPS shows the pro- posed approach can provide a reliable and flexible online target tracking for nautical radar application. It is also shown that, in the scenario of strong sea clutter, the proposed approach outperforms the PF-only based approach and the classical track- ing approach which combines order-statistics (OS) CFAR processing and the Kalman filter.