Radar Multiframe Detection in a Complicated Multitarget Environment

In this article, we develop a multiframe detection architecture relying on the generalized likelihood ratio test (GLRT) to address complicated multitarget detection and tracking in radar systems. The commonly used and unrealistic assumptions made in the previous works, where targets are sufficiently far apart from each other and do not occupy the same measurement cell (only for high-resolution radars), can be relaxed. Specifically, at the design stage, we take full account of the energy integration of real targets over multiframe measurements and the accurate data association of multitarget tracks for arbitrarily located targets. To solve the above challenges, a new detection statistic including the amplitude likelihood and the transition cost of a search path is derived. Then, we come up with two detection schemes for both known and unknown number of targets consisting of the log-likelihood ratio, an offsetting term about the transition cost and a penalty term accounting for the unknown number of targets. Finally, a fast implementation for the above detectors is investigated to trade detection and tracking performance for a lower computational complexity. Interestingly, the proposed architectures can not only improve detection performance for multiple dim targets through noncoherent integration, but also suppress tracks swap or mispairing among multiple frames when the targets are close. Numerical results and tests with real radar data for various multitarget scenarios are provided to demonstrate the effectiveness of the proposed algorithms.

Automated summary

In this article, a multiframe detection architecture based on GLRT is proposed to handle complex multitarget detection and tracking in radar systems. Previous assumptions about target separation and single measurement cell occupancy are relaxed. The architecture considers energy integration of real targets and accurate data association for arbitrarily located targets. A novel detection statistic is derived, and two detection schemes for known and unknown number of targets are developed. The proposed architectures improve detection performance for multiple dim targets and suppress tracks swap or mispairing. Numerical results and real radar data tests verify the effectiveness of the algorithms.