Multiple-Kernelized-Correlation-Filter-Based Track-Before-Detect Algorithm for Tracking Weak and Extended Target in Marine Radar Systems
Yi Zhou, Hang Su, Shuai Tian, Xiaoming Liu, Jidong Suo
Abstract
This article addresses the problem of tracking weak and extended targets in the clutter for marine radar systems. In the proposal, multiple kernelized correlation filters (MKCFs) incorporate a low-threshold constant false alarm rate and segmentation model under the principle of multiframe track before detect (MF-TBD). By setting the similarity score of the correlation filter as the test statistic, the maximization of the integration over frames can be solved by the essentially exhaustive searching of the MKCF. Since the correlation filter measures the similarity of the intensity distributions in the extended templates, the proposed method can differentiate the weak target from the surrounding clutter and other targets at a close range. Compared to other amplitude-based MF-TBD methods, it performs better in detecting and tracking extended targets in the real-radar data with heavy clutter caused by floating sea ice and the simulations with Rayleigh and $K$-distributed sea clutters.