Multi-target Track Before Detect with Labeled Random Finite Set and Adaptive Correlation Filtering

In Track-Before-Detect (TBD), the aim is to jointly estimate the number of tracks and their states from low signal-to-noise ratio (SNR) images. This is a challenging problem due to the unknown and time varying number of targets as well as the nonlinearity and size of the image data. A good balance between tractability and fidelity is important in the design of the measurement model for such trackers. In this paper, we transform the raw images into predetection images via adaptive correlation filtering, then apply an efficient labeled random finite set tracking filter for image data. Moreover, instead of using a particle implementation, we use an unscented transformation implementation which is computationally efficient and does not suffer from particle depletion. Numerical studies using realistic radar-based TBD scenarios are presented to verify the efficiency of the proposed solution.